A Dynamic Structural Analysis of BonusBased Compensation Plans
*
Doug J. Chung, Yale School of Management
Thomas Steenburgh, Harvard Business School
K. Sudhir, Yale School of Management
March 2011
.
*
The authors thank the marketing seminar participants at Arizona, Chicago, Emory, HKUST, IIM, Bangalore,
ISB, Maryland, Rochester, Stanford, UNC and Yale and participants at the Yale IO lunch, 2010 Choice
Symposium at Key Largo, 2009 AMA Doctoral Consortium, 2009 UTDFORMS conference and the Yale
Doctoral Workshop for helpful comments and suggestions.
Do Bonuses Enhance Sales Productivity?
A Dynamic Structural Analysis of BonusBased Compensation Plans
Abstract
We estimate a dynamic structural model of sales force response to a bonus based
compensation plan. The paper has two main methodological innovations: First, we
implement empirically the method proposed by Arcidiacono and Miller (2010) to
accommodate unobserved latent class heterogeneity with a computationally light twostep
estimator. Second, the bonus setting helps estimate discount factors in a dynamic structural
model using field data. This is because, quarterly and annual bonuses help generate the
instruments necessary to identify both discount factors in a hyperbolic discounting model.
Substantively, the paper sheds insights on how different elements of the compensation plan
enhance productivity. We find evidence that: (1) bonuses enhance productivity across all
segments; (2) overachievement commissions help sustain the high productivity of the best
performers even after attaining quotas; and (3) quarterly bonuses help to improve
performance of the weak performers by serving as pacers to keep the sales force on track to
achieve their annual sales quotas. We also find clear evidence of hyperbolic discounting by
salespeople.
1
1. Introduction
Personal selling is one of the most important elements of the marketing mix,
especially in the context of B2B firms. An estimated 20 million people work as
salespersons in the United States (Albers et al. 2008). Sales force costs average about 10%
of sales revenues and as much as 40% of sales revenues for certain industries (Albers et al.
forthcoming). In the aggregate, U.S. firms spent over $800 billion on sales forces in 2006, a
number that is three times larger than advertising spending (Zoltners, Sinha and Lorimer
2008).
Marketing researchers routinely create response models for marketing mix
instruments such as price, sales promotion and advertising. Metaanalysis of various
research studies estimate that the sales force expenditure elasticity is about 0.35 (Albers et al.,
2008) , relative to about 0.22 for advertising (Assmus, Farley and Lehmann 1983) and about 
2.62 for price (Bijmolt et al. 2005). While relative sales force expenditure elasticity is
useful in determining the relative effectiveness of different instruments in the marketing mix,
they give us little insight on how to design a sales force compensation plan, which is widely
understood to be the primary tool by which firms can induce the sales force to exert the
optimal levels of effort and thus to optimize the use of sales force expenditures.
A compensation plan can consist of many components: salary, commissions, and
bonuses on achieving a certain threshold of performance called quotas. Figure 1 shows a
variety of compensation plans that include combinations of these components. According to
Joseph and Kalwani (1998), only about 25% of firms use a pure commissionbased plan; the
rest used some form of quotas. As per the Incentive Practices Research Study (2008) by ZS
Associates, 73%, 85% and 89% in the pharma/biotech, medical devices and high tech
industries respectively uses quota based compensation.
This paper has two substantive goals: First, to gain insight on how a firm should
design its compensation plan. Specifically, should a firm offer quotas and bonuses in
addition to commissions? Despite the ubiquity of quotabased compensation, there is
considerable controversy in the theoretical (e.g., Holmstrom and Milgrom 1987; Lal and
Srinivasan 1993) and empirical literature (Oyer 1998; Steenburgh 2008) about the
effectiveness of quotas and bonuses relative to straight linear commission plans. Our paper
sheds light on this controversy by estimating a dynamic structural model of how the sales
2
force responds to alternative compensation instruments and specific levels of commission
rates, quotas and bonus levels.
Second, what should be the frequency of bonuses? Should one use a monthly,
quarterly or annual bonus? Should one use a quarterly bonus in addition to an annual bonus?
In the education literature, researchers have argued that frequent testing leads to better
performance outcomes (BangertDrowns et al. 1991). Can quarterly quotas serve a similar
role to improve outcomes? Like in the education literature, where frequent exams keep
students prepared for the comprehensive final exam; frequent quotabonus plans may serve as
a mechanism to keep the sales force motivated to perform in the shortrun well enough to be
in striking distance of the overall annual performance quota.
Methodologically, the paper offers two key innovations. First, we empirically
implement unobserved heterogeneity in a latent class framework within a computationally
light two step conditional choice probability (CCP) framework to estimate the dynamic
structural model. Though the use of two step estimation approaches have recently gained
popularity (Hotz and Miller 1993; Bajari, Benkard and Levin 2007), due to ease of
computation relative to traditional nested fixed point estimation approaches (e.g., Rust 1987),
their use in empirical applications have been limited by their inability to accommodate
unobserved heterogeneity. Arcidiacono and Miller (2010) propose an approach that allows
accommodation of latent class heterogeneity within the twostep estimation framework.
However, there are few empirical applications of this approach. To the best of our
knowledge, ours is among the first empirical papers applying the Arcidiacono and Miller
approach to account for unobserved heterogeneity in the twostep dynamic structural
estimation framework.
2
Second, and of great importance to the dynamic structural modeling literature, we
estimate rather than assume discount factors. It is wellknown in the literature on dynamic
structural models that discount factors cannot be identified in standard applications because
there are no instruments that provide exclusion restrictions across current and future period
payoffs (Rust 1994). Hence the standard approach is to assume discount factors. In
contrast to this, in our application, we have natural instruments in the form of bonuses: in
2
Two concurrent working papers that have implemented this approach in economics are Finger (2008) and
Beauchamp (2010).
3
nonbonus periods, bonuses should have no impact on current period payoffs, but only on
future payoffs. This enables us to estimate discount factors from the data. Further, the
psychology literature has shown strong evidence of hyperbolic discounting or present bias,
where in contrast to the constant exponential discounting (Samuelson 1937), researchers have
shown evidence of ‘hyperbolic discounting’ (Thaler 1981). The key idea of hyperbolic
discounting is that individuals discount the immediate future from the present more than they
do for the same time interval starting at a future date and hence a declining discount rate.
The most frequently given example is the preference reversal shown between two delayed
rewards. An individual may prefer 100$ today to 120$ in a year but may also prefer 120$ in
two years to 100$ in a year.
Hyperbolic discounting is typically mathematically represented using the following
quasihyperbolic discount function at time t: D(t)=βδ
t
(Phelps and Pollak (1968), Elster
(1979), Laibson (1997, 1998). Hence we need to estimate two discount parameters: a short
run present bias factor (β<1) and a longterm discount factor (δ). When β=1, the model
reduces to the single parameter exponential discounting model. Fortunately, the presence of
bonuses at different frequencies (quarterly and annual) provides us instruments necessary to
identify both the discount parameters.
There are three specific modeling and estimation challenges in the structural
estimation of response to compensation plans, especially those with quotas and bonuses.
First, in a typical structural model, one observes the agent's action in response to the firm's
action. For example in a consumer response model, one observes consumers’ choices in
response to the firm's choice of marketing mix such as price, advertising or sales promotion.
In contrast, for a sales force response model, one does not observe the actions of the sales
force, i.e., the exerted effort. One only observes the outcome of the agent's effort, i.e., sales,
which is correlated (but not a onetoone mapping) with effort. Hence one has to make an
inference about the agent's action (effort) that leads to sales from the observed realized sales.
This requires some modeling assumptions on the link between sales and effort.
3
3
The issue has parallels in empirical channel response models. For example, Sudhir (2001) makes an
inference about manufacturer actions (wholesale prices) from the observed retail price and sales to infer
competition between manufacturers.
4
A second challenge is that unlike marketing mix variables that change over time, the
compensation plan remains stationary over at least a year. How can one identify how the
sales force will respond to a compensation plan, when there is no variability in the plan?
Here we draw on an empirical insight from Steenburgh (2008) that can help identify the sales
force response, when the compensation plan involves payments for reaching quotas. The
insight is that in any given period, a sales agent's optimal effort depends on her state: how
close the person is to achieving her quota. A sales agent may find it optimal to reduce effort
when she is close or very far from achieving quota, but may stretch herself to reach the quota,
when she has a moderate chance of achieving this quota. This implies that the optimal level
of effort (and therefore sales) would vary from period to period as a function of the agent's
state (distance to quota).
A third issue follows from the discussion of the second. While quotas enable
identification of sales force response, it also induces intertemporal dynamics in optimal sales
force response behavior. An agent has to be concerned not just with the current payoff by
expending effort, but the future payoff that she can obtain by being in a more favorable state
that can facilitate in obtaining a bonus. This implies that the estimated structural model
needs to account for forwardlooking behavior on the part of sales agents. This requires a
dynamic structural model.
We estimate the dynamic structural model of sales force response to various features
of the compensation plan using sales force output and compensation data from a Fortune 500
firm that sells office durable goods. This firm used Plan F in figure 1b. In addition, the
quotas and bonuses are provided at two different frequencies: quarterly and annual. As the
compensation structure of the focal firm features almost all dimensions in typically used
compensation plans, we observe how the sales force responds to these different dimensions of
the plan. This rich plan provides us two key benefits: First, the presence of bonuses allows
us to identify and estimate discount factors, which would otherwise be impossible. Further,
since the bonuses are at different frequencies, we are able to estimate discount factors
(present bias factor and longterm discount factor) in a hyperbolic discounting model.
Second, even though theoretically one can perform counterfactuals of any type of
compensation plan if we can estimate structural parameters (other than discount factors) for a
sales person with a less rich compensation plan, an analyst or manager should have greater
5
faith in the counterfactuals, based on parameters that were estimated from observed responses
to different elements of the compensation plan.
We conclude the introduction with a numerical example that clarifies (1) how bonuses
can efficiently stretch sales people to exert more effort for a given level of compensation; (2)
how a person’s distance to quota can induce heterogeneity in effort. Let the utility function
of the salesperson that trades off effort (e) and income from sales (s), who has sold S units at
the beginning of the new period be:
0(s, e, S) = ue
2
+ is + B · 1
{s+S>Q]
where –d is the disutility parameter and r is the commission rate (d>0, r>0) and B is the
bonus obtained upon reaching quota (Q). For simplicity, assume a onetoonemapping
between sales and effort, i.e., s=e.
We begin with the case where all agents that are identical in their distance to quota in
the current period, specifically where S=0, i.e., agents have sold nothing thus far. Consider
the pure commission case with no bonus where d=1, r=10 and B=0. In this case, the optimal
effort is e*=5. For the bonus case, assume Q=10, and B=30. In this case, the optimal
effort is higher at e*=10. The compensation cost to the firm is $130. To achieve the same
level of effort from a pure commission plan, the commission rate r has to be increased to 20
and would cost the firm $200. Figure 2a illustrates these results graphically. Thus the
quotabonus plan is more efficient in obtaining the same level of sales.
Next let us consider the scenario where d=2, r=10, Q=10 and B=30, where agents are
different in their distance to quotas such as S=0 (far away from quota), S=5 (moderately close
to quota) and S=7 (close to quota). Figure 2b shows the results graphically, with maximum
effort of e* when S=5, relative to when S=3 or S=7.
For an intertemporal model, where bonuses occur every few periods, the decision to
stretch to obtain a bonus would depend on how close to bonus the sales person would be in
the bonus period. But this also means that the sales person needs to look forward
dynamically when exerting effort in earlier periods in order to be in a "good" state to reach
the quota and receive a bonus in the bonus periods.
6
The rest of the paper is organized as follows. Section 2 discusses the related
literature. Section 3 summarizes the institutional details of the firm and data used for the
empirical analysis. We present the model and the estimation methodology in sections 4 and
5. Section 6 discusses the estimation results and the counterfactual analysis. Section 7
concludes.
2. Related Literature
The literature review is in two parts: We begin with the discussion of the literature
relating to the substantive issue of the role of nonlinearities such as quotas and bonuses in
compensation plans. Following this, we discuss the empirical literature on structural
estimation of worker productivity.
In the theoretical literature, Basu et al. (1985) apply the principal agent framework of
Holmstrom (1979) and demonstrate that a combination of salary plus commission (usually
nonlinear with respect to sales) will be optimal. Rao (1990) also shows a similar result on
the optimality of nonlinear compensation plans. However, Holmstrom and Milgrom (1987)
and Lal and Srinivasan (1993) question the need for nonlinear compensation schemes.
Using the specific assumptions of linear exponential utility and normal errors (LEN) they
show that a linear commission incentive scheme can achieve the best possible outcomes for
the firm.
Yet, why do most firms have quota based compensation plans? Why are
compensation plans nonlinear? Raju and Srinivasan (1996) suggest that even though a
commission over quota plan may not be theoretically optimal, they provide the best
compromise between efficiency and ease of implementation. Others argue that quota based
plans offer high powered incentives that can motivate salespeople to work harder (e.g.,
Darmon 1997). Park (1995) and Kim (1997) demonstrate that a quotabonus plan may lead
to the firstbest outcome, but in their framework, quotabonus plan is just one of many
possible plans that lead to first best outcomes. Oyer (2000) shows that when participation
constraints are not binding, a quotabonus plan with linear commissions beyond quotas can
be uniquely optimal because it can concentrate the compensation in the region of effort where
the marginal revenue from effort minus the cost of compensation is maximized.
7
There is limited empirical work addressing this issue. Based on an analysis of
aggregate sales across different industries in different quarters, Oyer (1998) concludes that
the negative effects of quota based plans encouraging sales people to maneuver the timing of
orders are greater than the benefits obtained from more effort. Steenburgh (2008) questions
whether aggregate data can be used to reach this conclusion. Using individual sales
performance data from the same firm used in this study (utilizes compensation plan F in
figure 1b), he finds that the net improvement in revenues from effort dominates the
inefficiencies induced by intertemporal dynamic considerations.
Our work is related to several static structural models of worker behavior such as
Ferrall and Shearer (1999) and Paarsch and Shearer (2000), who endogenize the optimal
contract choice of the firm, given linear contracts. In contrast to these papers, we seek to
understand the response to nonlinear incentives, which require us to model the dynamic
response of sales agents. However, we do not model the contract choice, because we do not
have data on selection across contracts.
Copeland and Monnet (2008) estimate a dynamic structural model of worker
productivity in a checksorting environment with nonlinear incentives; unlike sales force
productivity, there is limited unobserved uncertainty in check sorting productivity. Much of
the variation in productivity here can be explained by observed characteristics such as
machine breakdowns etc.
A contemporaneous paper by Misra and Nair (2009) on sales force compensation is
closest to this paper in methods and substantive context. In terms of methodology, both
papers use the twostep estimation technique; however our paper innovates on two key
dimensions. First, we accommodate latent class heterogeneity within the twostep
estimation framework—an issue that has been an econometric challenge for the literature for
close to two decades. Misra and Nair sidestep the unobserved heterogeneity issue by
estimating each sales person’s utility function separately.
4
Second, unlike Misra and Nair,
who assume discount factors, we contribute to the broader dynamic structural modeling
4
This approach to accommodate heterogeneity is similar to the estimation of individual level utility functions in
conjoint analysis or scanner panel data, by using a large number of observations related to a particular individual.
Further, the approach requires that sales people will exert effort equally across all customersan assumption
they show is valid in their data, but unlikely to hold in general. Our method of using latent classes works in the
more common situation where there are limited observations per individual.
8
literature by estimating discount factors (specifically hyperbolic discount factors) using field
data. Two characteristics of the data facilitate identification of discount factors. First,
bonus payoffs are in the future and are excluded from current payoffs of the nonbonus
months. This helps us estimate discount factors. Second, bonuses are at two different
temporal distances, quarterly and annual, and as a result this helps us estimate both the
present bias factor and the longterm discount factor in a hyperbolic discounting model.
The key substantive difference is that we focus on the value of quotas with bonuses
(plan F in figure 1b), while Misra and Nair consider the role of quotas with floors and
ceilings on commissions (plan D in figure 1b). Misra and Nair conclude that quotas reduce
performance. This is because of two characteristics of their quotas: First, the quota ceiling
(beyond which sales people receive zero additional compensation) limits the effort of the
most productive sales people who would normally have exceeded the ceiling. Second, the
company followed an explicit policy of ratcheting quotas based on past productivity, which
reduced incentives of sales people to work hard in any given period, because it makes future
rewards more difficult to attain due to ratcheted higher quotas. In contrast, we find that
quotas coupled with bonuses enhance performance. In the plan we consider, the company
offers extra overachievement commissions for exceeding quotas and use a group quota
updating procedure that minimizes ratcheting effects. Thus the two papers offer
complementary perspectives that enhance our understanding about how quotas impact
performance.
3. Institutional Details and ModelFree Evidence of Dynamics
We begin by describing the details of the compensation plan. Then we provide
modelfree evidence to highlight the salient characteristics of the data to be modeled.
3.1. The Compensation Plan
The focal firm under study is a highly regarded multinational Fortune 500 company
that sells durable office products primarily using its own direct sales force. It also uses an
indirect sales force through “rep” firms. These “rep” firms do not compete with the direct
sales force.
5
Further, each sales agent is given “exclusive” territories; therefore, only one
5
Such rep firms are the focus of Jiang and Palmatier (2010).
9
sales agent receives credit for a unit of sales and the firm traditionally does not encourage
group work or team cooperation among the sales force.
Our analysis focuses on sales performance data from 348 sales people during the
three year period 19992001. The firm’s compensation structure follows the pattern in Plan
F of Figure 1b and the details of the compensation schedule for the period of analysis are
described in Table 1. Every month, sales people receive a fixed monthly salary and
commissions per volume of sales generated in that month. There are no caps on revenues
for which an agent could obtain commissions or overachievement commissions. In the first
three quarters, a quarterly lumpsum bonus is given when each of the quarterly quotas are met.
And at the end of the year (i.e., end of the fourth quarter) an annual lumpsum bonus is paid if
the annual quota is met and an overachievement commission is given for excess revenues
beyond the annual quota. In building its annual and quarterly quota, the company allocates
monthly quotas to its sales force, though these are not used for compensation.
The single most important element, in terms of performance evaluation within the
firm, is the annual quota; i.e., the firm views a salesperson as having a successful year if the
annual quota is met. Overall, for a salesperson that meets all quotas, the salary component
will be roughly 30% of total compensation.
3.2 Model Free Analysis
We consider three features of the data that can inform model development. First, we
consider the nature of seasonality in the data. Second, we look at the evidence of forward
looking behavior induced by bonuses and hence the need to develop a dynamic model.
Finally, we test for the possibility of sales substitution across quarters by sales agents.
3.2.1 Seasonality
The descriptive statistics of the data is in Table 2. Figure 3a shows the average
revenue generated for each calendar month by the sales force in the study. Sales is
relatively high at the end of first three quarters (March, June, September) and much higher in
the last quarter (December). At first glance, this suggests that sales agents are highly
responsive to quotas and bonuses.
10
But could the higher sales at the end of quarters be due to higher demand during
these periods, rather than just due to bonuses? How could we control for the effects of
seasonality? We were able to obtain aggregate sales revenue data from the indirect sales
force who is compensated purely on a commission basis. Since these salespeople were
purely compensated on commissions, they can serve as a proxy for seasonality in demand.
Figure 3b shows the monthly revenue of the indirect sales force. The shape of the monthly
revenue for the indirect sales force shows distinct seasonality suggesting demand fluctuates
from period to period independent of compensation. We therefore use this variable to
account for demand seasonality.
Why should we expect sales to be so seasonal and coincide by quarter? Given that
the focal firm’s products are a B2B business selling discretionary good whose timing of
purchase are also flexible, one possibility is that spending may expand during the month in
which the fiscal year ends for its customers due to accounting procedures employed by
clients. Figure 3c shows the distribution of fiscal yearends across months for the year 2000
(the midpoint of our sample). Indeed, we see that over 66% of the firms have a December
fiscal yearend, which might explain a significant boost in sales during that month. End of
quarter months also have peaks relative to other months, but the peaks are not as large as for
December. Given that we have more direct proxies for seasonality in the form of indirect
sales force revenues, we do not use the fiscal yearend information in our analysis.
3.2.2: Forward Looking Behavior induced by quarterly and annual bonuses
We begin by showing scatter plots and the best fitting nonparametric smoothed
polynomial (and its 95% confidence interval) of sales revenues normalized by monthly
allocated quotas in the quarterly bonus months (March, June, September, December) against
percentage of quota attained by the previous month in figure 4a. For March, June and
September, the x axis is the percentage of quarterly quota completed (%QQ), while for
December, the x axis is the percentage of annual quota completed (%AQ). The vertical line
shows the %QQ and %AQ at which the salespeople on average achieve their monthly
allocated quotas.
Two key elements stand out from the graphs. First, across the board one does not
see a massive reduction in effort when salespeople get closer to achieving their quota. This
must be partly because of the overachievement commission rate. Second, we can detect
11
evidence of forward looking behavior. Early in the year (March and June), salespeople
achieve their monthly targets, even at low levels of %QQ, while later in the year, they
achieve their monthly targets only at fairly high levels of %QQ or %AQ. In March and
June, even when there is little chance of achieving the quarterly quota, the sales person puts
in effort to achieve quota. On its own, this might simply mean that salespeople are trying to
obtain commissions. However, when seen in tandem with the fact that sales agents in
similar states in September or December do not seek to accomplish their targets, one can infer
that the sales person is forward looking. Early in the year, even if they are below targets,
they still have hopes of receiving the large annual bonus by working hard. However, later in
the year, such chances becomes less likely and sales people respond by reducing their effort.
This is clear evidence of forward looking behavior. This also suggests an instrument for
estimating discount factors. Annual bonuses should have little impact on current payoffs in
March, June or September, but only on future payoffs. Hence distance to annual quotas can
serve as an instrument for estimating discount factors.
The next set of graphs presented in Figure 4b, shows the same relationship in the pre
bonus months (February, May, August and November). Here a new characteristic stands
out. In the early months, February, May and even August, at all levels of %QQ, the
salesperson on average sells above the monthly allocated quota. This is because hard work
(and some luck in the form of positive shocks) may give a reasonable chance of attaining the
smaller quarterly targets. However, in November, only at a very high level of %AQ, does
the salesperson sell above the monthly allocated quota, because one has a very limited chance
of making up the large gap in just two months.
This suggests again that in the prequarterly bonus months, the quarterly bonus in the
future does have an impact on behavior, even though it does not have an immediate impact
on the current payoff; again indicating forward looking behavior. However, the differential
way in which the sales person responds to temporally different bonuses at any given point in
time, provides an opportunity to identify the two discount factors in hyperbolic discounting: a
longterm discount factor and a present bias factor.
This preliminary evidence also suggests a natural question for managers. Should the
large annual bonus be split into a quarterly bonus (as in other months) and an annual bonus?
This can prevent sales people from giving up in November, even if they do not have a chance
12
of reaching the annual quota. At the same time, the cost of such a quarterly quota would be
that early in the year, the incentive to stretch after reaching quarterly quotas would be
reduced. How these two issues tradeoff is an empirical question, which we subsequently
address in the counterfactual analysis.
3.2.3. Sales substitution across months
One possibility is that sales people giving up at the end of the quarter may be doing so
to increase the odds of hitting quotas in subsequent quarters by simply not booking the sales
in the current quarter. If this were true, then one should see a negative linkage between
sales in month t and month t+1; and especially between the last month of a quarter and first
month of the next quarter.
To test this, we report the results of a regression where sales in month t is regressed
against sales in month t1 in table 3. Specifically to see if there might be any borrowing
effects from the last month of a bonus period to the first period of the next bonus period we
include a separate variable for the first month of each quarter. We also include individual
level fixed effects. We do not find any evidence of substitution across months. Further,
we do not find any significant coefficient for the first month of quarter (as in Model 2),
suggesting little substitution across quarters. The first month of quarter also does not have
any significant impact even if we separated the effect for people who are “way off target” in
the last month and have therefore the greatest incentives to postpone purchases, as seen in
Model 3. We defined “way off target” as those whose previous quarter sales were less than
50% of their quota. The results were robust and did not vary with alternative definitions of
“way off target”. We therefore do not model substitution across quarters.
4. Model
Based on the modelfree evidence, we build a dynamic model of sales force response
to the quotabased compensation scheme. The timing of the model is as follows:
1. At the beginning of each year, firm chooses the annual compensation plan.
2. Each month, agents observe their current state and exert effort in a dynamically
optimal manner.
3. An idiosyncratic sales shock is realized; the shock plus agent's effort determines the
agent's realized sales for the period. Agent receives compensation.
13
4. The realized sales of the current period affect the agent's state of the next period.
Steps 23 are repeated each month until the end of the year and Steps 13 are repeated
over the years.
We now describe the model in five parts: (i) the compensation plan (ii) the sales
agent’s utility function (iii) the state transitions (iv) effort as a function of state variables and
(v) the optimal effort choice by the sales agent.
4. 1. Sales Response Model
Estimating a sales response model with respect to effort poses a challenge because
effort is not observed. This implies that typical endogeneity issues in choice of marketing
mix such as price and advertising, where the marketing mix choice is correlated with certain
market characteristics that are observable to the decision maker, but unobservable to the
researcher, cannot be accounted for. To appreciate this, consider a hypothetical scenario,
where one can observe effort. Then we would estimate the following two equation model 
one for sales (S), and one for effort (e) as follows:
it it it
S e + z +ε
D
it it
β γ ν = +
it it
e h(z , ; , )
E
it it i
ν μ ω = ϒ +
Here
D
it
z are sales shifters, and
E
it
z are effort shifters,
it
ν are unobserved shocks to
the researchers that can be observed by the sales agents.
it
ε and
it
ω are the sales and
productivity shocks respectively. h is a function (that solves a static or dynamic model) that
maps the consumer observable shifters (
E
it
z
and
it
ν ) to effort. Since
it
ν
is not observed to
the researchers, it is necessary to assume that the h function is monotonic in
E
it
z
and
it
ν ,
even when effort is observed. But when effort is unobserved, one needs to solve for the
effort function only on researcher observable characteristics. This requires us to assume
away
it
ν and
it
ω , i.e., we need to rule out potential endogeneity effects and productivity
shocks in the application.
Give the above discussion, we model the sales revenue function (S
It
) for salesperson i
at time t in two parts: (1) a base level of sales independent of effort, parameterized by
14
demand shifters ( z
D
it
) and (2) sales induced due to effort (e
It
), which is a function of effort
shifters that include both territory and salesperson characteristics ( z
E
it
).
S
It
= f( z
D
it
) + e
It
( z
E
it
) + ε
It
… (1)
where ε
It
is an additive sales revenue shock that is not anticipated by the salesperson when
choosing effort levels.
As discussed earlier, the market potential varies across territories, but also has a time
varying seasonal component. To account for the crosssectional variation in market
potential, we use annual quota from the previous year (
1
AQ
iy −
). To account for seasonality
of demand across months, we use the indirect sales revenues for month t, (IS
t
). We also
include an interaction between the two variables to account for the possibility that seasonality
will have a larger impact on larger territories.
For the effort shifters in e
it
, we use the following variables: Given that effort is a
function of demand shifters, we include both
−1
AQ
y
and IS
t
in z
E
it
. As discussed in the
motivation, the salesperson’s state with respect to achieving quota will have an impact on the
effort they expend. We therefore use the cumulative percentage of quarterly and annual
quota completed till time t (%QQ
it
, %AQ
it
) as variables that affect effort. In addition, we
allow a timeinvariant salesperson specific variable, tenure with the firm (Tenure) to
moderate the level of effort.
Note that unlike the demand shifter function f, which is common across all
salespeople, the effort function will vary across salespeople. Specifically, we allow for
salespeople to belong to one of multiple discrete segments, hence these effort functions will
be estimated at the segment level. We estimate the effort function nonparametrically, by
using Chebyshev polynomials of the variables described above.
4. 2 Compensation Plan
The compensation plan has three components. They are: (1) the monthly salary w
it
,
(ii) endof quarter bonus, B
iqt
for achieving the corresponding quarterly quota Q
iqt
, and end of
year bonus B
iyt
for achieving the corresponding annual quota Q
iyt
(3) commission rate r
it
per
dollar worth of sales and an overachievement commission rate, r
it
′ given at the end of the year
15
for sales over and above the annual quota for each individual i at time t. We represent the
compensation plan for a salesperson i by the vector ψ
it
={w
it
, Q
iqt
, Q
iyt
, B
iqt
, B
iyt
, r
it
, r
it
′}.
4. 3 Sales person’s perperiod utility
In each period t, sales person i receives positive utility of wealth W
it
earned based on
realized sales and a disutility C(e
it
; θ
i
) from exerting effort e
it
. Thus the utility function is
defined as:
0(e
It
, S
It
; ψ
I
, θ
I
, γ
I
) = EW(S
It
; ψ
I
)]  γ
I
vaiW(S
It
; ψ
I
)]  C(e
It
; θ
I
)
6
where γ
i
and θ
i
are each the risk aversion and disutility parameters respectively for
salesperson i.
Given the sales levels, and the compensation plan, the wealth for individual i, W
it
can
be computed. W
it
arises from four components, the per period salary component w
it
, the
lumpsum bonus component B
it
, the commission component C
it
, and the overachievement
commission component OC
it
. The detailed expressions of wealth is as follows,
W
It
= w
It
+ B
It
+C
It
+ OC
It
B
It
= 1
qt
· 1
_z
¡1t
+
s
¡t
(c
¡t
(z
¡t
E
),z
¡t
D
; u
¡
)+ε
¡t
Q
¡qt
> 1_
· B
qt
+ 1
yt
· 1
_z
¡2t
+
s
¡t
(c
¡t
(z
¡t
E
),z
¡t
D
; u
¡
)+s
¡t
Q
¡yt
> 1_
· B
yt
C
It
= i
It
· (s
It
(e
It
(z
It
E
), z
It
D
; α
I
) + ε
It
)
OC
It
= 1
yt
· 1
_z
¡2t
+
s
¡t
(c
¡t
(z
¡t
E
),z
¡t
D
; u
¡
)+ε
¡t
Q
¡yt
> 1_
· r′
It
· [z
I2t
· Q
Iyt
+ s
It
(e
It
(z
It
E
), z
It
D
; α
I
)+ε
It
 Q
Iyt
¸
where z
i1t
and z
i2t
are the percentage of quarterly and annual quotas completed respectively
by salesperson i until time t. 1
qt
and 1
yt
are indicators for whether time t is a quarterly or
annual bonus period.
6
In the case of the CARA utility function (exponential utility function) with normal errors and a linear
compensation plan, this functional form represents the certainty equivalent utility of the agent. Here we
consider the utility function to be a second order approximation to a general concave utility function.
16
In our empirical analysis, we use a quadratic functional form for the disutility
function; specifically, C(e; θ
I
) = θ
I
c
2
. Thus the set of structural parameters of the
salesperson that needs to be estimated are Ω
I
= (θ
I
, γ
I
).
4. 4 State Variables
As discussed, the nonlinearity of the compensation scheme with quotas and bonuses
introduces dynamics into the sales agent's behavior because there is an additional tradeoff
between the disutility of effort today and a higher probability of lumpsum bonus and
overachievement commissions tomorrow. To incorporate the dynamics of the model we
consider the following stochastic state variables, the percentage of annual quota completed,
the percentage of quarterly quota completed. These state variables evolve as follows:
1. Percentage of quarterly quota completed (%AQ)
z
I1t
= _
u, if t is stait of quaiteily quota peiiou
z
I1(t1)
+
S
I(t1)
Q
Iqt
, othei wise
2. Percentage of annual quota completed (%QQ)
z
I2t
= _
u, if t is stait of annual quota peiiou
z
I2(t1)
+
S
I(t1)
Q
Iyt
, othei wise
Other state variables would include time varying demand shifter, indirect sales, and territory
characteristics, of which we use previous year’s annual quota. Naturally, the time varying
indirect sales is a onetoone mapping to period type and hence includes the information
about different periods. We also observe individual characteristic, specifically tenure with
the focal firm (τ), and therefore use it as an individual state variable that impacts effort.
These state variables are collected in a state vector z
It
E
= z
I1t
, z
I2t
, IS
t
, AQ
I(y1)
, τ
I
.
4.5 Optimal Choice of Effort
Given the parameters of the compensation scheme ψ, and the state variables and their
transitions, each sales agent would choose an effort level conditional on her states to
maximize the discounted stream of expected future utility flow. Alternatively, if this value
function is below the reservation wage, the salesperson may choose to leave the firm.
17
The stream of utility flow, under the optimal effort policy function, conditional on
staying at the firm, and the behavioral notion of quasihyperbolic discounting can be
represented by a value function,
v(z; ψ, Ω) = max
e
U(e, z; ψ, Ω) +βδEj max
c
U(e, z
i
; ψ, Ω) +δEjmax
c
U(e, z′′; ψ, Ω) +·[ [
or differently put as,
v(z; ψ, Ω) = max
c
U(e, z; ψ, Ω) +βδEv
8
(z
i
; ψ, Ω)]
v
8
(z′; ψ, Ω) = max
e
U(e, z′; ψ, Ω) +δEv
8
(z′′; ψ, Ω)]
where Ω is the primitives or the structural parameters of the underlying utility function,
specifically the disutility parameter θ and the risk aversion parameter γ. β and δ are the
discount parameters. The longterm discount factor is δ, the shortterm discount factor is βδ,
where β<1 represents the present bias in hyperbolic discounting. If β=1, the model reduces
to a single parameter exponential discount model. The expectation of the value function is
taken with respect to both the present and future sales shocks.
5. Estimation
Traditionally, the nested fixedpoint algorithm (NFXP) developed by Rust (1987) is
used to estimate dynamic models. However, NFXP estimators are computationally
burdensome as one has to solve the dynamic program numerically over each guess of the
parameter space for every iteration. The twostep estimation first introduced by Hotz and
Miller (1993) and extended by Bajari, Benkard, and Levin (2007) can serve to reduce the
computation burden. In this approach, the model estimation proceeds in two steps. In the
first step, the conditional choice probabilities of choosing a certain action as a function of
state variables are estimated in a flexible nonparametric manner. Then, in the second step,
these conditional choice probabilities are used to estimate the structural parameters of the
sales agent's utility function. For this approach to work, it is critical that the conditional
choice probabilities are estimated accurately in the first step.
Until recently, it was believed that the accurate estimation of conditional choice
probabilities for an agent is impractical when there is unobserved heterogeneity.
18
Arcidiacono and Miller (2010) have proposed an EM–Algorithm based approach to
accommodate unobserved heterogeneity in the first step of the two step estimation procedure.
We provide one of the first applications of this approach – illustrating the empirical validity
of the approach in practical applications. We now discuss the details of the two step
estimation procedure.
5. 1 Step 1
In this step, we need to estimate a flexible nonparametric mapping between
observable states and actions of the sales person; this requires a nonparametric model of the
monthly effort function e
I
(z
It
E
), that links effort and state in equation (1). We model the
effort function nonparametrically as a combination of basis functions of the state variables.
Thus the nonparametric effort function is:
e
It
= ρ
ℓ
(z
It
E
)
L
ℓ=1
· λ
Iℓ
… (2)
Where the ℓ
th
basis function is ρ
ℓ
(z
It
E
). In this application, the ℓ
th
basis function is the ℓ
th
order Chebyshev polynomial.
From equations (1) – (2) we have the following sales response function to estimate.
S
It
= f(z
It
D
; α
I
) +ρ
ℓ
(z
It
E
) · λ
Iℓ
L
ℓ=1
+ ε
It
For z
it
D
,
which is a subset of z
it
E
(from now on refer to as z
it
), we use two variables: (1) lagged
annual quota for salesperson i, (2) the revenues of the indirect sales force. We use the direct
linear effect of these variables to control for cross sectional variations of territory
characteristics and temporal variations in monthly seasonality. The interactions effects of
these variables with the other state variables goes into the polynomial function in (2).
The lagged annual quota takes into account general territory characteristics that are
likely to be generated independent of effort, i.e., market size. The revenues from the
indirect sales force capture market seasonality, independent of the nonlinear nature of the
compensation plan. We assume that the revenue shocks (ε
It
), come from a normal
distribution that are i.i.d. across time.
19
If one could estimate the sales response and effort response function at the level of
each individual, we can simply obtain the individual level parameters of the effort and sales
policy function by maximizing the log likelihood of the sample such as
Θ
´
I
= aigmax log _L
I
_S
It
 f(z
It
D
; α
I
)  ρ
ℓ
(z
It
E
) · λ
Iℓ
L
ℓ=1
__
T
t=1
, … (S)
where the vector Θ
´
I
= {α
I
, λ
I
, σ
I
] contains the set of parameters of the sales response and
effort policy functions and the distribution of sales shocks, where
L
I
(ε) =
1
o
¡
√2¬
e

1
2
_
s
o
¡
]
2
… (4)
We accommodate unobserved heterogeneity by allowing for discrete segments.
Assume that sales person i belongs to one of K segments, k ∈ {1,…,K} with segment
probabilities q
i
={q
i1
,…, q
iK
}. Let the population probability of being in segment k be π
k
.
Let L(S
It
z
It
, k; Θ
k
) be the likelihood of individual i's sales being S
it
at time t, conditional
on the observables z
it
, and the unobservable segment k, given segment parameters Θ
k
. Then
the likelihood of observing sales history S
i
over the time period t=1…T, given the observable
history z
i
, and the unobservable segment k is given by:
L
k
(S
I
 z
I
; Θ
k
, π
k
) = _q
Ik
L
Ikt
T
t=1
… (S)
where L
Ikt
= L(S
It
 z
It
, k; Θ
k
). As noted earlier we assume the distribution of the revenue
shocks to be normally distributed and hence use the normal likelihood for equation (5) as in
equation (4). The parameter Θ
k
= {α
k
, λ
k
, σ
k
] is the vector of segment level parameters of
the sales response and effort policy function where each λ
k
is the parameters that index the
effort policy for segment k and σ
k
is parameter for the distribution of the revenue shocks for
segment k.
By summing over all of the unobserved states k∈{1,…,K}, we obtain the overall
likelihood of individual i:
L(S
I
 z
I
; Θ, π) = q
Ik
L
k
(
K
k=1
S
I
 z
I
; Θ
k
, π
k
)
20
and hence the loglikelihood over the N sample of individuals becomes
log (L(S
I
 z
I
; Θ, π))
N
I=1
= log (q
Ik
_L
Ikt
T
t=1
K
k=1
)
N
I=1
… (6)
Directly maximizing the loglikelihood in (6) is computationally infeasible because the
function is not additively separable so we take the approach of Arcidiacono and Jones (2003)
and Arcidiacono and Miller (2010) to iteratively maximize the expected loglikelihood in
equation (7)
q
Ik
logL(S
It

T
t=1
z
It
k; Θ
k
) … (7)
K
k=1
N
I=1
where q
ik
is formally defined below as the probability that individual i is of segment type k
given parameters values Θ, segment probabilities π, and conditional on all of the observed
data of individual i.
Pi(kS
I
, z
I
; Θ, π) = q
Ik
(S
I
, z
I
; Θ, π) =
I
k
(S
I
 z
I
; Θ
k
, π
k
)
L(S
I
 z
I
; Θ, π)
… (8)
The iterative process is as follows.
We start with an initial guess of the parameters Θ
0
and π
0
. Natural candidate for
such starting values would be to obtain the parameters from a model without unobserved
heterogeneity and slightly perturbing those values.
7
Given the parameters {Θ
m
, π
m
} from
the m
th
iteration, the update of the (m+1)
th
iteration is as follows
(a) Compute q
ik
(m+1)
using equation (8) with Θ
m
and π
m
(b) Obtain Θ
m
by maximizing (7) evaluated at q
ik
(m+1)
(c) Update π
(m+1)
by taking the average over the sample such that
π
k
(m+1)
=
1
N
q
Ik
(m+1)
N
I=1
We would iterate (a) – (c) till convergence.
7
We started the initial values from one tenth of the standard error from the parameter values obtain from a
single segment model. The initial values of the segment probabilities were set equally across segments.
21
For the basis functions in the effort policy, we use Chebyshev polynomials of state
variables to approximate effort.
8
As a result, we obtain the vector of parameters that index
these basis functions λ′s, the vector of parameters for the sales policy α′s, and the parameters
of the revenue shocks σ’s
for each segment k. Also we obtain the population segment
probabilities for each segment. More formally,
Θ
´
= {Θ
1
¯
, …, Θ
K
¯
= (α
k
´, λ
k
¯
, σ
k
´)]
π¯ = {π
1
´, …, π
K
´]
Therefore, we obtain the sales revenue function S
`
(. ) and effort policy function
e´(. ) for each segment.
5. 2 Step 2
The key idea of the twostep estimation is that in the 1
st
stage we observe the agent’s
optimal actions. Using these observed optimal actions we are able to construct estimates of
the value function, which enables us to estimate the primitives of the model that rationalize
these optimal actions.
Let the value function of a representative agent at state z that follows an action
profile e, conditional on the compensation plan ψ, the sales profile S and the primitives of the
utility function Ω be represented as
v(z; e; ψ, S, Ω) = E_B(t)0(e(z
t
), z
t
, ε
t
; Ω)
T
t=0
_z
0
= z; ψ, S, Ω_ … (9)
where B(t) = _
1, if t = u
βδ
t
, othei wise
is the hyperbolic discount factor, and the expectation
operator would be over the present and future sales shock ε
t
.
Using the estimated sales and effort policy function and the distribution of the sales
shocks in the first stage, we are able to forward simulate the actions of sales agents to obtain
the estimate of the value function. The detailed simulation procedure is as follows.
8
For reference, see “Numerical Methods in Economics”, Kenneth L. Judd, MIT Press, 1998.
22
(a) From initial state of z
t
calculate the optimal actions as e(z
t
)
(b) Draw sales shock ε
t
from f(ε)
(c) Update state z
t+1
using the realized sales s(e(z
t
))+ ε
t
(d) Repeat (a) – (c) until t=T
By averaging the sum of the discounted stream of utility flow over multiple simulated paths
we can get the estimate of the value function v
¯
(z; e(z); ψ, S, Ω).
9
Let e
s
(z) be any deviation policy from a set of feasible policies that is not identical to
the optimal policy and, by using the same simulation method proposed above, let the
corresponding estimate of the value function be called the suboptimal value function
v
¯
(z; e
s
(z); ψ, S, Ω). Since e(z) by definition is the effort policy and thus at an optimum,
then any deviations from this policy rule would generate value functions of less or equal
value to that of the optimal level.
Let us define the difference in the two value functions as,
Q(v; ψ, S, Ω) = v(z; e(z); ψ, S, Ω) v(z; e
s
(z); ψ, S, Ω)
where v e V denotes a particular {z, e
s
(z)} combination.
10
Then if e(x) is the optimal
policy, the function Q(v;ψ,Ω) would always have value of greater or equal to zero. Thus
our estimate of the underlying structural parameters Ω would satisfy,
Ω
´
= argmin _(min{Q(v; ψ, S, Ω), 0])
2
dH(v)
where H(v) is the distribution over the set V of inequalities. Our empirical counterpart to
Q(v; ψ, S, Ω) would be Q
¯
(v; ψ, S
`
, Ω) = v
¯
(z; e´(z); ψ, S
`
, Ω) v
¯
(z; e´
s
(z); ψ, S
`
, Ω) and as a
result our estimates of the structural parameters are obtained from minimizing the objective
function in equation (10).
11
9
For each segment, we drew four hundred simulation draws over each period and computed the value functions.
10
As indicated in Bajari, Benkard, and Levin (2007), there are multiple ways to draw these suboptimal policy
rules. Although the method of selecting a particular perturbation will have implications for efficiency the only
requirement necessary for consistency is that the distribution of these perturbations has sufficient support to
yield identification. We chose to draw a deviation policy from a normal distribution with mean zero and
quarter of the variance from the revenue shock distribution, i.e. e
s
(z)=e(z)+η
11
We drew two hundred deviation strategies to construct the objective function and hence N
I
=200.
23
1
N
I
(minQ
¯
(v
j
; ψ, S
`
, Ω), u)
2
N
I
j=1
… (1u)
The above procedure is performed for each segment with the segment specific effort
policies obtained in Step1. This allows us to estimate the structural parameters for each
segment.
12
5.3. Identification
We provide a brief and informal discussion of identification. Realized sales is a
function of effort and additive sales shocks. Given multiple observations of sales at
different states, we can separately identify nonparametrically the density of sales shocks and
a deterministic function of effort. We make a parametric assumption about a strictly
monotonic relationship between sales and effort because it is not possible to identify this
relationship nonparametrically.
Further, we assume a deterministic (but highly flexible) relationship between effort
and observable states (percentage of quarterly and annual quotas achieved by previous month
and the demand shifters) at the level of each segment. Thus variation in sales (which is
monotonically linked to effort) as a function of these state over time helps identify the effort
disutility parameter. The risk parameter is identified by differences in response to variations
in wealth levels over time.
The discount factor is typically not identified in dynamic choice models because
instruments affect both the current utility and future utility (Rust 1994; Magnac and Thesmar
2002). In our model, the state variables (%QQ and %AQ) serve as identifying instruments.
The choice of effort in nonbonus periods has two distinguishable wealth effects; (1) the
increase in current utility through the commissions and (2) the increase in future utility
through bonuses. Without these future bonuses, a sales person would just simply solve the
myopic first order condition every period to decide on the optimal level of effort. However,
with the future bonuses, a sales person can choose to increase effort to increase the likelihood
12
We also estimated a model with a second set of moment inequalities to reflect the participation constraint that
employees continued to work at a firm because they at least obtained a reservation value (normalized to zero);
i.e., min(v(z; e(z); ψ, S, Ω),0). The constraint was nonbinding and did not impact our estimates.
24
of bonus, conditional on the state variables and depending on how much she values the future,
which by definition is the discount factor.
Further, having two different future bonuses, one in the near future (quarterly
bonuses) and the other in the distant future (annual bonuses and overachievement
commissions) with different state variables affecting them, help us identify the hyperbolic
discount parameters. Indeed, the modelfree evidence in Figure 4 shows different responses
to annual and quarterly bonuses.
6. Results
We first report the first stage estimates of the demand shifters and effort policy
function for the sales response model; then we report estimates of structural parameters of
sales agents' utility functions from the second stage estimation. In the second stage
estimation, we perform a grid search over the discount parameters (betas and deltas). We
then perform several counterfactual simulations to address the substantive questions we seek
to answer.
6. 1 First Stage Estimates
The parameter estimates for the demand shifters in the sales response function is
reported in Table 4a. We find that only the interaction term between lagged annual quota
and indirect sales revenue are statistically significant. Thus larger markets tend to have a
bigger sales multiplier independent of effort in high demand periods.
We estimate segment level effort policy functions by estimating the nonparametric
relationship between sales and state variables through Chebyshev polynomials of the state
variables. We estimated up to fourth order Chebyshev polynomials with alternative number
of segments and choose the best fitting model based on the Bayesian Information Criterion
(BIC). Uniformly, three segment models had the best fit. The estimates of the best fitting
Chebyshev polynomial function (ρ
n
indicates the nth order Chebyshev polynomial) and the
standard deviations of the revenue shocks for each segment are reported in Table 4b and 4c.
As the coefficients associated with the Chebyshev polynomials have no intuitive meaning, for
intuition, we show graphs of the effort policy function for the three segments as a function of
percentage annual quota (%AQ) for March (end of first quarter) and December (end of year)
25
in Figure 5. %AQ is normalized across sales agents, such that 1 implies at quota and 0.9
indicates 10% below quota and 1.1 indicates 10% above quota.
Table 5 shows the share of the three segments and their descriptive characteristics.
Segment 2 is the largest with a share of 47%; Segments 1 and 3 have shares of 32% and 21%
respectively. The average tenure with the firm is not very different across segments at
approximately 12 years. Segment 3 has the highest annual quotas, followed by Segment 2
and Segment 1. Interestingly, Segments 2 and 3 with larger quotas achieve their quota
targets more often than Segment 1 which has trouble meeting quota.
Figure 5a shows the Segment 3 exerts the most effort and is the most productive
segment, and Segment 1 exerts the least effort and is the least productive segment. This is
consistent with the allocated quotas and percentage of time quotas are achieved in Table 3.
We also see a positive relationship between exerted effort and %AQ for all months shown.
As for %QQ, we see an increasing but concave relationship in March implying that once a
sales person is way above the quarterly quota she starts to gradually slow down. Given that
the average states in March for each segment were 0.55, 0.58, 0.62, respectively, many sales
people are not in a position to slow down. Effort in December does not fall off even if the
sales person has already reached or exceed quota (%AQ>1), likely due to the
overachievement commissions that incentivize sales people to keep exerting effort even after
achieving quota. Our results are consistent with Steenburgh (2008), who finds that sales
people “give up” when far away from achieving quota, such as for all segments in our case,
but do not slow down much once quota is reached.
Figure 5b shows the effect of tenure on effort for all segments. Sales people in
segment 2 and 3 initially increase effort with experience, but this tapers off with time. This
is probably due to the fact that in the early years of their careers, they want to work hard not
only for monetary payments from increased wages but also other intangible incentives such
as promotions or transfers to better job titles. However, after a certain number of years in
the same job, these career concerns do not matter as much and the effort levels tend to taper
off. Interestingly, Segment 1, the lowest productivity segment, does not gain in productivity
from experience.
6.2 Discount Factor
26
We performed a grid search over the set of discount parameters in steps of 0.01 for
delta and 0.1 for beta. Table 6 presents the mean absolute percentage errors (MAPE)
associated with each set of hyperbolic parameters where a beta equals one represents
exponential discounting. A beta of 0.8 and a delta of 0.92 has the lowest MAPE.
13
Thus
our estimates show a distinct present bias in that β <1.
Frederick, Loewenstein and O’Donoghue (2002) have a comprehensive summary of
the estimated discount factors from previous studies. The summary shows that the
estimated discount factors vary extensively ranging from as low as a mere 0.02 to no
discounting at all with a discount factor of 1. For purely monetary values, the estimated
discount factor seems rather low. But as Frederick, Loewenstein and O’Donoghue (2002)
point out, for behavioral aspects such as pain and thus in our case effort, the discount factors
tend to be low and hence our estimates appear reasonable.
6. 3 SecondStage Structural Parameter Estimates
The first column of Table 7 reports the structural parameter estimates of the sales
agent's utility function for a forward looking sales person, consistent with the model we
developed earlier. Overall, the disutility parameters for all three segments are negative and
significant. These estimates are consistent with the effort policy functions estimated in the
first stage. Segment 3, which produces the greatest sales on average, has the lowest
disutility for effort. Segment 1, which has the lowest sales, has the greatest disutility. The
risk aversion coefficients for all segments are insignificant showing no direct evidence of risk
aversion by the sales agents. This may be because in the range of incomes earned by the
sales force, risk aversion is not a serious concern. The estimated model fits the observed
sales revenue data reasonably well with a MAPE of 10.7%.
6. 4 Assessing the value of a dynamic structural model
How important is it to account for dynamics to model salesperson behavior? If
salespeople are behaving in a dynamically optimal manner, not accounting for dynamics
would bias the structural parameters. In a static model, any effort would be attributed to
current payoff, not accounting for the large future bonuses. This will underestimate the
13
We tested for finer granularity around beta for 0.8 and found that 0.8 is optimal.
27
salesperson disutility parameters and overestimate the effects of compensation on
productivity.
The second column of Table 7 shows the estimates of the structural model without
forward looking dynamics – discount factor set to zero. As predicted, the disutility
parameters are smaller in magnitude relative to the forward looking model for all segments.
For Segment 3 the bias is as large as 22% relative to the dynamic model estimate. The
myopic model also has a poorer fit: a MAPE of 18.8% relative to the MAPE of 10.7% for the
dynamic model.
We next compare the revenue and effort predictions between the dynamic and
myopic models. To isolate the effects of forward looking behavior, we use the structural
disutility parameter estimates from the dynamic model for both the forward looking and
myopic models, but set the discount parameters to zero for the myopic model. Figure 6
compares the predicted revenues and effort of the myopic and dynamic models. The
revenues are systematically lower for the myopic sales agent because the sales person does
not take into account the effect of future bonuses and overachievement commission in current
effort. The forward looking sales person anticipates that in an uncertain environment, there
is a chance of bad shocks later, which may prevent getting to the quota, so they prepare for
such a rainy day by working harder early on so that they are within striking target of quota
even if a bad sales shock occurred.
The effort graph in Figure 6 enables us to isolate out the sales revenue cyclicality and
focus on the differences in effort across dynamic and myopic agents. The myopic salesperson
exerts much more effort in the bonus period, but the forward looking sales person smoothes
effort over time, given the uncertainty in future demand shocks. The effort peak in the
bonus periods are not as pronounced for the dynamic consumer. The observed effort
smoothing is similar to consumption smoothing by forward looking consumers facing
uncertain incomes in the development economics literature.
6. 5 Counter–Factual Simulations
We now perform a series of counterfactual simulations that address the two sets of
substantive questions we wish to answer. First, we address the issue of how valuable
different components of the compensation plan are. The overall change in revenues under
28
the alternative conditions is reported in Table 8 and the effect by segment in Table 9.
Second, we assess compare the role of bonus frequency— how quarterly and annual bonuses
affect performance.
Value of Quotas and Bonuses
We compare changes in revenues and profits when the firm moves from the current
compensation plan to a pure commissiononly plan. We consider two cases: (1) where the
commission rate is the same as the current commission rate; and (2) a higher commission rate
is such that total compensation is exactly equal to the current compensation. We find that
the revenues are about 20.8% greater with the current compensation plan compared to a pure
commission plan. Interestingly, we find from Table 9 all segments suffer from substantially
poorer performance when quotas are removed and the firm shifts to a pure commission
scheme. Even when adjusting commission rates to be higher to make total compensation
identical to current levels, we find that revenues are about 4% higher.
Value of Overachievement Compensation
We compare changes in revenues and profits when the firm eliminates the over
achievement commission rate, which motivates sales people who are close to reaching their
quota to continue exerting effort. Overall revenues drop by 13.3% and even accounting for
the additional commission costs, profits are lower by about 2% (assuming gross margin of
33%).
Figure 7 plots the effort level of sales agents who met and didn’t meet the annual
quota, respectively. For those who met the annual quota, the effort level does not decline
even when close to the quota because of the overachievement commission. In contrast,
those who did not meet quota decrease effort towards the end of the year as they are unlikely
to meet quota and therefore overachievement commission has no impact on their earnings.
Thus overachievement commission provides the incentives for the most productive sales
people even if they have already met quota (or likely to meet quota). Not surprisingly Table
9 indicates that overachievement commissions have the most impact on Segments 2 and 3.
Value of Cumulative Annual Quota
29
Rather than have a cumulative annual quota, what would be the effect of replacing it
with just a fourth quarter quota? To study this, we remove the overachievement
commission (which is based on reaching the annual quota) and split the total bonus payments
across all four quarters. Overall, revenues drop by 15.8%. This decrease is greater than
the 13.3%, where we just dropped the overachievement commissions. Thus the cumulative
annual quota induces sales agents to exert greater effort and raise revenues by 2.5%.
We also consider the case where we split the annual quota into a quarterly bonus and
an annual bonus so that people do not “give up” in the last quarter when they are far away
from quota. While this did increase the effort in the last quarter, it reduced revenues overall
because sales people did not put in as much effort earlier in the year to be within striking
distance of annual quota, because it is not as large. Total revenues drop by 1%.
The Role of Quarterly versus Annual Quota
We next investigate the value of quarterly bonuses relative to annual bonuses.
Figure 8 shows the comparison of effort between the current plan and when quarterly bonuses
are eliminated and only an annual bonus is paid at the end of the year. Effort drops
consistently across the year when there are no quarterly quotas. Overall revenues fall by 5%.
Even in December, when there is the annual bonus on the table, revenue falls by 2% and
effort falls by 4%. Thus annual quotas and overachievement commissions have less of an
impact on yearend performance without quarterly bonuses. Why?
Not only does the quarterly quota induce sales agents to work harder in a given
quarter but it also helps them achieve the annual quota by giving the incentive to stay on
track of their annual goals. When the quarterly quota is removed, sales agents no longer
have as much incentive to work hard early on. But this lack of incentive leads them to be
farther away from the annual quota by December. Hence, annual quotas and over
achievement commission have little impact on effort as sales agents are more likely to give
up meeting quota.
The impact of quarterly bonuses also differs across the three segments of consumers.
Table 9 indicates that quarterly bonuses have relatively minimal impact on Segment 3, the
most productive segment, but very high impact on Segment 1. In effect, quarterly bonuses
30
are needed as pacers to the less productive sales people than for the most productive sales
people.
To the best of our knowledge, there has been no analysis todate on the role of bonus
frequency in enhancing sales productivity. There has been some descriptive work in the
education literature on how frequent testing affects academic performance (for an extensive
survey, see BangertDrowns et al. 1991) and some experimental work in the behavioral
psychology (Heath, Larrick and Wu, 1999). The basic idea is that achieving shortterm
goals make achieving longterm goals more feasible. Our analysis show that the short term
goals are more valuable to the least productive segment; i.e., in education terms it may imply
that weaker students gain more by periodic testing, relative to stronger students who would
study independent of exams.
7. Conclusion
Even though personal selling is a primary marketing mix tool for most B2B firms to
generate sales, there is little research on how the compensation plan used to motivate the sale
force affects performance. This paper developed and estimated a dynamic structural model
of sales force response to a compensation plan with many components: salary, commissions,
lumpsum bonus for achieving quotas, and different commission rates beyond achieving
quotas. Our analysis helped us assess the impact of (1) different components of
compensation and (2) the differential importance of periodic bonuses on performance on
different segments of sales people.
There has been a fair amount of controversy on the value of quotas and bonuses in
the literature. Overall, we find that the quotabonus scheme used by this firm increased
performance of the sales force by serving as stretch goals and pushing employees to
accomplish targets. Features such as overachievement compensation reduce the problems
associated with sales agents slacking off when they get close to achieving their quota.
Further, quarterly bonuses serve as a continuous evaluation scheme to keep sales
agents within striking targets of their annual quotas. In the absence of quarterly bonuses, a
failure in the early periods to accomplish targets caused agents to fall behind more often than
in the presence of quarterly bonuses. Thus, the quarterly bonus serves as a valuable sub
goal which helps the sales force stay on track in achieving their overall goal. A key finding
31
is that annual bonuses are not as effective for the sales force without quarterly bonuses for the
low productivity segments. For the most productive segments, quarterly quotas does not
seem to matter much. In summary, we find that bonuses have an overall positive impact on
all segments of salespeople. Overachievement commissions serve to increase performance
among the highest performers, while quarterly bonuses are most helpful to increase
performance among the weaker performers.
We use recent innovations in the twostep dynamic structural model estimation to
accommodate unobserved heterogeneity in sales force response. The approach is flexible,
yet computationally feasible with minimal additional burden compared to traditional twostep
methods. Features of our data allow us to separate seasonality in sales due to quotas as
opposed to underlying consumer demand seasonality. This enables us to get better estimates
of the response to compensation, because demand peaks that coincide with quota periods may
be wrongly interpreted as a byproduct of compensation.
We now discuss limitations of the paper, which provide promising avenues for future
research. First, effort tends to be multidimensional and one possibility is that quotas and
bonuses force people to focus on the effort that lead to final sales in bonus periods, while in
other periods, they may focus on earlier stages of the selling process. It is not possible to
identify such a multidimensional effort merely from the sales data as in this paper.
Nevertheless new data from CRM databases which track customer stages through the selling
process can help shed insight on this issue. This we believe is an exciting area for future
research.
Second, compensation contracts can serve as a selection mechanism to draw the right
type of sales people into the sales force. This paper does not address the selection issues.
By looking at a longer panel of sales people's performance, one can use attrition information
to address this issue. Looking at scenarios where contracts varied over time, can also shed
light on this problem. Papers that have looked at varying contracts over time typically have
focused on only contracts with linear commission rates (e.g., Paarsch and Shearer 1998).
One needs more work on scenarios with richer contracts.
Finally, Chan, Li and Pierce (2009) investigate the effects of peer effects on sales
performance in the presence of team based compensation in a reduced form analysis. It
would be useful to extend structural analysis to settings involving team based compensation.
32
In summary, this paper provides important insights on how the sales force responds
to a very rich compensation structure involving many components of compensation: salary,
commissions, quota and bonuses at quarterly and annual frequencies. How employees
respond to such rich compensation structures with bonuses, a reality at many firms, has not
been investigated at all in the literature. This paper illustrates a rigorous framework to
analyze this problem and obtains useful substantive insights. Nevertheless, the issues raised
above provide an interesting agenda for future work.
33
References
Albers, Sönke, Murali K. Mantrala, Srinath Gopalakrishna and Kissan Joseph, “Introduction:
Special Issue on Enhancing Sales Force Productivity”, Journal of Personal Selling and Sales
Management, 28, pp. 109113, 2008
Albers, Sonke, Murali K. Mantrala, and Shrihari Sridhar, “Personal Selling Elasticities: A
MetaAnalysis”, Journal of Marketing Research, forthcoming
Arcidiacono, Peter and Robert A. Miller, “CCP Estimation of Dynamic Discrete Choice
Models with Unobserved Heterogeneity”, Working paper, Duke University, 2010
Assmus, Gert, John U. Farley and Donald R. Lehmann, “How Advertising Affects Sales:
MetaAnalysis of Econometric Results”, Journal of Marketing Research, 21, pp. 6574, 1984
Bajari, Patrick, C. Lanier Benkard, and Jonathan Levin, “Estimating Dynamic Models of
Imperfect Competition”, Econometrica, 75(5), pp. 13311370, 2007
BangertDrowns, Robert L., James A. Kulik and ChenLin C. Kulik, “Effects of frequent
classroom testing”, Journal of Educational Research, 85, pp. 8999, 1991
Basu, Amyia K., Rajiv Lal, V. Srinivasan and Richard Staelin, “Salesforce Compensation
Plans: An Agency Theoretic Perspective”, Marketing Science, 8 (3), pp. 324342, 1985
Beauchamp, Andrew, “Abortion Supplier Dynamics”, Working Paper, Duke University, 2010
Bijmolt, Tammo H.A., Harald J. Van Heerde and Rik G.M. Pieters, “New Empirical
Generalizations on the Determinants of Price Elasticity”, Journal of Marketing Research 42,
pp. 141–56, 2005
Chan, Tat Y., Jia Li, and Lamar Pierce, “Competition and Peer Effects in Competing Sales
Teams”, Working Paper, Washington University, St. Louis, 2009
Copeland, Adam and Cyril Monnet, “The Welfare Effects of Incentive Schemes”, Review of
Economic Studies, 76, pp. 96113, 2009.
Darmon, Rene, “Selecting Appropriate Sales Quota Plan Structures and Quotasetting
Procedures”, Journal of Personal Selling and Sales Management, 17, pp. 116, 1997
34
Elster, Jon, “Ulysses and Sirens: Studies in Rationality and Irrationality” Cambridge
University Press, 1979
Ferrall, Christopher, and Bruce Shearer, “Incentives and Transactions Costs within the Firm:
Estimating an Agency Model Using Payroll Records," Review of Economic Studies, 66, pp.
309338, 1999
Finger, Stephen R., “Research and Development Competition in the Chemicals Industry”,
Working Paper, Duke University, 2008
Frederick, Shane, George Loewenstein and Ted O’Donoghue, “Time Discounting and Time
Preference: A Critical Review”, Journal of Economic Literature, 40, pp. 351401, 2002
Heath, Chip, Richard P. Larrick and George Wu, “Goals as Reference Points”, Cognitive
Psychology, 38, pp. 79109, 1999
Holmstrom, Bengt, “Moral Hazard and Observability”, Bell Journal of Economics, 10, pp.
7491, 1979
Holmstrom, Bengt and Paul Milgrom, “Aggregation and Linearity in the Provision of
Intertemporal Incentives”, Econometrica, 55, pp. 303328, 1987
Hotz, V. Joseph. and Robert A. Miller, “Conditional Choice Probabilities and the Estimation
of Dynamic Models”, Review of Economic Studies, 61, pp. 265289, 1993
Jiang, Renna and Robert W. Palmatier (2009) “Structural Estimation of a Moral Hazard
Model: An Application to Business Selling,” Working Paper, UC Davis.
Joseph, Kissan and Manohar Kalwani, “The Role of Bonus Pay in Sales force Compensation
Plans”, Industrial Marketing Management, 27, pp. 147159, 1998
Judd, Kenneth L., “Numerical Methods in Economics”, MIT Press, 1998
Kim, Son Ku, "Limited Liability and Bonus Contracts" Journal of Economics and
Management Strategy, 6, pp. 899913, 1997
Laibson, David, “Golden Eggs and Hyperbolic Discounting”, Quarterly Journal of
Economics, 112, pp. 443477, 1997
35
Laibson, David, “LifeCycle Consumption and Hyperbolic Discount Functions”, European
Economic Review, 42, pp. 861871, 1998
Lal, Rajiv and V. Srinivasan, ”Compensation Plans for Singleand Multiproduct Sales
forces: An Application of the Holmstrom – Milgrom Model”, Management Science, 39, pp.
777793, 1993
Magnac, Thierry and David Thesmar, “Identifying dynamic discrete choice models”,
Econometrica, 70, pp. 801816, 2002
Misra, Sanjog and Harikesh Nair, “Quota Dynamics and the Intertemporal Allocation of
Sales–Force Effort”, Working Paper, Stanford University, 2010
Oyer, Paul, “Fiscal Year Ends and Nonlinear Incentive Contracts: The Effect on Business
Seasonality”, Quarterly Journal of Economics, 113, pp. 149185, 1998
Oyer, Paul, “A Theory of Sales Quotas with Limited Liability and Rent Sharing”, Journal of
Labor Economics, 18, pp. 405426, 2000
Paarsch, Harry, J. and Bruce Shearer, ''The Response of Worker Effort to Piece Rates:
Evidence from the British Columbia TreePlanting Industry,'' Journal of Human Resources,
34, pp. 643667, 1999
Paarsch, Harry, J. and Bruce Shearer, ''Piece Rates, Fixed Wages, and Incentive Effects:
Statistical Evidence from Payroll Records”, International Economic Review, 40, pp. 5992,
2000
Park, EunSoo, “Incentive Contracting under Limited Liability”, Journal of Economics and
Management Strategy, 4, pp. 477490, 1995
Phelps, Edmund. S. and Robert Pollak, “On SecondBest National Saving and Game
Equilibrium Growth”, Review of Economic Studies 35, pp. 185199, 1968
Raju, Jagmohan S., and V. Srinivasan, “QuotaBased Compensation Plans for MultiTerritory
Heterogeneous SalesForces”, Management Science, 42, pp. 14541462, 1996
36
Rao, Ram C., “Compensating Heterogeneous Salesforces: Some Explicit Solutions”,
Marketing Science, 9, pp. 319342, 1990
Rust, John, “Optimal Replacement of GMC Bus Engines: An Empirical Model of Harold
Zurcher”, Econometrica, 55, pp. 9991033, 1987
Rust, John, “Structural Estimation of Markov Decision Processes”, in Handbook of
Econometrics, Volume 4, ed. by R.E. Engle and D. McFadden, Amsterdam: ElsevierNorth
Holland, Chapter 14, pp. 30823139, 1994
Samuelson, Paul, “A Note on Measurement of Utility”, Review of Economic Studies, 4, pp.
155161, 1937
Steenburgh, Thomas, “Effort or Timing: The Effect of LumpSum Bonuses”, Quantitative
Marketing and Economics, 6, pp. 235256, 2008
Sudhir, K., “Structural Analysis of Manufacturer Pricing in the Presence of Strategic
Retailer”, Marketing Science, Summer, pp. 244264, 2001
Thaler, Richard H., “Some Empirical Evidence on Dynamic Inconsistency”, Economic
Letters, 8, pp. 201206, 1982
Zoltners, Andris A., Prabhakant Sinha and Sally E. Lorimer, “Sales Force Effectiveness: A
Framework for Researchers and Practitioners”, Journal of Personal Selling and Sales
Management, Special Issue on Enhancing Sales Force Productivity, 28, pp. 115131, 2008
37
Table 1: Firm's Compensation Plan
Type Description
Payment
period
Quarterly Bonus
$1500 Awarded if quarterly revenue exceeds quarterly
quota
Mar, Jun, Sep
Annual Bonus $4000 Awarded if annual revenue exceeds annual quota Dec
Base
Commission
About 1.5%
*
paid in proportion to the revenue
generated each month
Every month
Overachievement
Commission
About 3%
*
paid in proportion to the total cumulative
revenue surpassing the annual quota
Dec
*
These numbers are approximate for confidentiality reasons.
Table 2: Descriptive statistics – Sales force under study
Amount
% Achieving
Quota
Size 348 
Average Salary
$3,585 ‐
(USD)
Average tenure
11.8 
(years)
Average Q1 quota
232.4 51.1
(000'USD)
Average Q2 quota
374.2 49.8
(000'USD)
Average Q3 quota
397.1 42.8
(000'USD)
Average fullyear quota
1,639.3 49.9
(000'USD)
38
Table 3: Testing for Sales Substitution Across Months
Model 1 Model 2 Model 3
Last Month Sales
0.302
***
(0.014)
0.188
***
(0.013)
0.188
***
(0.014)
Qtr 1
st
Month *Last Month Sales
0.007
(0.013)
0.023
*
(0.013)
Qtr 1
st
Month*Last Month Sales*
“Way off Target Last Qtr”
0.029
*
(0.015)
Qtr 1
st
Month*Last Month Sales*
“Not Way off Target”
0.004
(0.024)
Monthly Allocated Quota
0.565
***
(0.021)
0.566
***
(0.021)
Indirect Sales
14.063
***
(2.130)
14.15
***
(2.229)
Sales person Fixed Effects Yes Yes Yes
***
: p<0.01
*
: p<0.1
Table 4a: Parameter Estimates – Sales Response
Lagged annual quota
0.002
(0.005)
Indirect sales
6.735
(5.554)
Indirect sales*Lagged annual quota
0.022***
(0.003)
***: p<0.01
39
Table 4b: Parameter Estimates – Effort Policy
Seg1 Seg2 Seg3
ρ
0
18.63 74.23** 263.00***
(26.09) (32.74) (89.28)
ρ
1
(z
1
)
141.10*** 146.42*** 312.36**
(44.32) (55.41) (137.85)
ρ
2
(z
1
)
23.54 21.67 148.39***
(17.09) (22.37) (63.23)
ρ
3
(z
1
)
6.41 6.98 15.87
(3.94) (5.14) (13.09)
ρ
1
(z
2
)
102.26*** 241.73*** 458.59***
(38.00) (44.93) (117.20)
ρ
2
(z
2
)
22.13 62.91*** 87.28
(18.67) (21.89) (55.50)
ρ
3
(z
2
)
8.09* 11.34*** 15.90
(4.39) (4.54) (10.79)
ρ
1
(z
1
)*ρ
1
(z
2
)
114.52 31.98 110.00
(75.28) (82.95) (181.61)
ρ
1
(z
1
)*ρ
2
(z
2
)
16.60 0.27 113.83***
(19.77) (18.40) (40.33)
ρ
2
(z
1
)*ρ
1
(z
2
)
9.07 10.14 86.81***
(14.63) (17.08) (36.85)
ρ
1
(z
1
)*ρ
1
(IS)
20.00** 8.12 42.42
(8.95) (11.31) (29.62)
ρ
1
(z
2
)*ρ
1
(IS)
1.11 16.76*** 20.10***
(2.68) (3.18) (8.53)
ρ
1
(z
1
)*ρ
1
(z
2
)*ρ
1
(IS)
9.52 43.91*** 89.11**
(17.62) (18.88) (44.87)
ρ
1
(z
1
)*ρ
1
(AQlag)
0.04** 0.03** 0.07***
(0.02) (0.02) (0.03)
ρ
1
(z
2
)*ρ
1
(AQlag)
0.01 0.06*** 0.02
(0.01) (0.01) (0.01)
ρ
1
(z
1
)*ρ
1
(z
2
)*ρ
1
(AQlag)
0.04 0.05*** 0.02
(0.04) (0.03) (0.04)
ρ
1
(T)
0.53 9.00*** 18.53***
(1.38) (2.14) (6.90)
ρ
2
(T)
0.05 0.27*** 0.52**
(0.05) (0.07) (0.26)
ρ
3
(T)
0.001 0.002*** 0.004
(0.000) (0.001) (0.003)
***: p<0.01, **: p<0.05, *: p<0.1
40
Table 4c: Revenue Shock Distribution – Standard Deviation
seg1 seg2 seg3
Sigma
81.61 143.72 279.35
Table 5: Descriptive Characteristics of Segments
segment1 segment2 segment3
Share
0.32 0.47 0.21
Tenure*
11.5 12.2 11.5
Achieve quarterly quota  Q1
0.46 0.54 0.58
Achieve quarterly quota  Q2
0.38 0.55 0.62
Achieve quarterly quota  Q3
0.31 0.49 0.53
Achieve annual quota
0.30 0.57 0.64
Average annual quota**
1,201.4 1,615.7 2,363.7
Average December revenue**
130.2 273.0 559.1
*Tenure is measured in years **Average quotas and revenues are indicated in USD(K)
Table 6: Optimal Discount Factor – Model Fit
Mean Absolute Percentage Error by Discount Factors
0.9 0.91 0.92 0.93 0.94 0.95 0.96 0.97 0.98 0.99
0.4 0.1336 0.1347 0.1367 0.1378 0.1391 0.1382 0.1402 0.1428 0.1450 0.1475
0.5 0.1249 0.1258 0.1243 0.1270 0.1255 0.1300 0.1266 0.1300 0.1287 0.1318
0.6 0.1167 0.1181 0.1193 0.1203 0.1219 0.1200 0.1207 0.1167 0.1198 0.1241
0.7 0.1103 0.1149 0.1134 0.1124 0.1107 0.1127 0.1140 0.1202 0.1220 0.1242
0.8 0.1104 0.1084 0.1074 0.1098 0.1100 0.1132 0.1149 0.1150 0.1175 0.1172
0.9 0.1121 0.1108 0.1101 0.1117 0.1148 0.1145 0.1180 0.1185 0.1192 0.1203
1 0.1109 0.1098 0.1103 0.1168 0.1181 0.1182 0.1167 0.1223 0.1251 0.1328
δ
β
41
Table 7: Parameter Estimates
With Forward Looking
Behavior
Without Forward
Looking Behavior
Segment 1
Disutility 0.240 0.199
(0.005) (0.006)
Risk Aversion 0.0001 0.0001
(0.0020) (0.0003)
Segment 2
Disutility 0.119 0.092
(0.038) (0.005)
Risk Aversion 0.0001 0.0001
(0.0005) (0.0005)
Segment 3
Disutility 0.061 0.048
(0.002) (0.002)
Risk Aversion 0.0000 0.0000
(0.0002) (0.0004)
***p<0.01
Table 8: Impact of Alternative Bonus Plans on Sales Revenues
Counterfactual Change in `Revenues
1a. Only Pure Commissions 20.8%
1b. Only Pure Commissions (adjusted to equal payout with bonus) 3.8%
2a. No Bonus (Only Commissions + Overachievement Commission) 9.3%
2b. No Bonus (Commissions adjusted to equal payout with bonus) 1.5%
3. No overachievement commissions 13.3%
4a. Cumulative Annual Quota replaced with quarterly quota 4.2%
4b. Annual Bonus split into Quarterly and Annual Bonus 1.0%
5a. Remove quarterly bonus 4.6%
42
Table 9: Impact of Alternative Bonus Plans on Sales Revenues by Segment
% decrease from different components Seg1 Seg2 Seg3
Pure commission 17.9% 21.0% 21.4%
Without overachievement 7.0% 12.6% 17.1%
Without quarterly bonus 10.0% 4.5% 2.0%
Figure 1: Plots of Incentive Compensation Schemes
E
a
r
n
i
n
g
s
Sales
Plan B: Pure Bonus
E
a
r
n
i
n
g
s
Sales
Plan C: Commission at Quota
E
a
r
n
i
n
g
s
Sales
Plan E: Commission+Bonus
E
a
r
n
i
n
g
s
Sales
Plan F: Commission+Bonus
+Overachievement Commission
E
a
r
n
i
n
g
s
Sales
Plan D: Commission with
Floor and Ceiling
E
a
r
n
i
n
g
s
Sales
Plan A: Pure Commission
43
Figure 2a: How Quotas and Bonus Serve as stretch goals
Figure 2b: Effort as a Function of Distance to Quotas
Figure 3: Sales / IndirectSales / Fiscal YearEnds
50
0
50
0 2 4 6 8 10
d=1, r=10,B=0,
e*=5
50
0
50
0 2 4 6 8 10
d=1,r=10,B=30
e*=10
0
200
0 2 4 6 8 10
d=1, r=20,B=0
e*=10
150
100
50
0
50
0 2 4 6 8 10
d=2, r=10, B=30
S=0, e*=2.5
100
50
0
50
0 2 4 6 8 10
d=2, r=10, B=30
S=5, e*=5
100
50
0
50
100
0 2 4 6 8 10
d=2, r=10, B=30
S=7, e*=3
0
50
100
150
200
250
300
J
a
n
F
e
b
M
a
r
A
p
r
M
a
y
J
u
n
J
u
l
A
u
g
S
e
p
O
c
t
N
o
v
D
e
c
U
S
D
(
K
)
a) Average Sales
0
0.5
1
1.5
2
2.5
3
3.5
J
a
n
F
e
b
M
a
r
A
p
r
M
a
y
J
u
n
J
u
l
A
u
g
S
e
p
O
c
t
N
o
v
D
e
c
N
o
r
m
a
l
i
z
e
d
S
a
l
e
s
b) Sales ‐ Indirect Sales Force
0
10
20
30
40
50
60
70
80
J
a
n
F
e
b
M
a
r
A
p
r
M
a
y
J
u
n
J
u
l
A
u
g
S
e
p
O
c
t
N
o
v
D
e
c
P
e
r
c
e
n
t
a
g
e
c) Percentage of Fiscal year‐ends
44
Figure 4a: Sales and Percentage Quota Achieved – Bonus Months
Figure 4b: Sales and Percentage Quota Achieved – PreBonus Months
0
1
2
3
M
a
r
S
a
l
e
s
/
M
a
r
Q
u
o
t
a
0 .2 .4 .6 .8 1 1.2 1.4
% Quarterly Quota Sold by Feb
kernel = epanechnikov, degree = 0, bandwidth = .18, pwidth = .27
0
1
2
3
J
u
n
S
a
l
e
s
/
J
u
n
Q
u
o
t
a
0 .2 .4 .6 .8 1 1.2 1.4
% Quarterly Quota Sold by May
kernel = epanechnikov, degree = 0, bandwidth = .33, pwidth = .5
0
1
2
3
S
e
p
S
a
l
e
s
/
S
e
p
Q
u
o
t
a
0 .2 .4 .6 .8 1 1.2 1.4
% Quarterly Quota Sold by Aug
kernel = epanechnikov, degree = 0, bandwidth = .23, pwidth = .35
0
1
2
3
D
e
c
S
a
l
e
s
/
D
e
c
Q
u
o
t
a
0 .2 .4 .6 .8 1 1.2 1.4
% Annual Quota Sold by Nov
kernel = epanechnikov, degree = 0, bandwidth = .17, pwidth = .26
0
1
2
3
F
e
b
S
a
l
e
s
/
F
e
b
Q
u
o
t
a
0 .2 .4 .6 .8 1 1.2
% Quarterly Quota Sold by Jan
kernel = epanechnikov, degree = 0, bandwidth = .16, pwidth = .23
0
1
2
3
M
a
y
S
a
l
e
s
/
M
a
y
Q
u
o
t
a
0 .2 .4 .6 .8 1 1.2
% Quarterly Quota Sold by Apr
kernel = epanechnikov, degree = 0, bandwidth = .17, pwidth = .25
0
1
2
3
A
u
g
S
a
l
e
s
/
A
u
g
Q
u
o
t
a
0 .2 .4 .6 .8 1 1.2
% Quarterly Quota Sold by July
kernel = epanechnikov, degree = 0, bandwidth = .21, pwidth = .31
0
1
2
3
N
o
v
S
a
l
e
s
/
N
o
v
Q
u
o
t
a
0 .2 .4 .6 .8 1 1.2
% Annual Quota Sold by Oct
kernel = epanechnikov, degree = 0, bandwidth = .2, pwidth = .29
45
Figure 5a: Effort Policy by Segment as a Function of % Quota
Figure 5b: The Effect of Tenure on Effort
Figure 6: Simulated Revenue & Effort– Static vs. Dynamic

50
100
150
200
250
0
.
0
0
0
.
0
5
0
.
1
0
0
.
1
5
0
.
2
0
0
.
2
5
0
.
3
0
0
.
3
5
0
.
4
0
0
.
4
5
0
.
5
0
0
.
5
5
0
.
6
0
0
.
6
5
0
.
7
0
0
.
7
5
0
.
8
0
E
f
f
o
r
t
% QQ
March

50
100
150
200
250
300
350
400
0
.
5
0
0
.
5
5
0
.
6
0
0
.
6
5
0
.
7
0
0
.
7
5
0
.
8
0
0
.
8
5
0
.
9
0
0
.
9
5
1
.
0
0
1
.
0
5
1
.
1
0
1
.
1
5
1
.
2
0
1
.
2
5
1
.
3
0
E
f
f
o
r
t
% AQ
December
(20)

20
40
60
80
100
120
1 2 3 4 5 6 7 8 9 10 11 12 13
E
f
f
o
r
t
Tenure (years)
The Effect of Tenure
46
Figure 7: Effect of Overachievement Commission
Figure 8: Effect of Quarterly Quotas

50
100
150
200
250
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
U
S
D
(
K
)
a. Revenue
Dynamic Myopic

20
40
60
80
100
120
140
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
U
S
D
(
K
)
b. Effort
Dynamic Myopic
30
50
70
90
110
130
150
J
a
n
F
e
b
M
a
r
A
p
r
M
a
y
J
u
n
J
u
l
A
u
g
S
e
p
O
c
t
N
o
v
D
e
c
U
S
D
(
K
)
Met quota Didn't meet quota
60
70
80
90
100
110
120
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
U
S
D
(
K
)
Current plan Without quarterly bonus
Do Bonuses Enhance Sales Productivity? A Dynamic Structural Analysis of BonusBased Compensation Plans
Abstract We estimate a dynamic structural model of sales force response to a bonus based compensation plan. The paper has two main methodological innovations: First, we implement empirically the method proposed by Arcidiacono and Miller (2010) to accommodate unobserved latent class heterogeneity with a computationally light twostep estimator. Second, the bonus setting helps estimate discount factors in a dynamic structural model using field data. This is because, quarterly and annual bonuses help generate the instruments necessary to identify both discount factors in a hyperbolic discounting model. Substantively, the paper sheds insights on how different elements of the compensation plan enhance productivity. We find evidence that: (1) bonuses enhance productivity across all segments; (2) overachievement commissions help sustain the high productivity of the best performers even after attaining quotas; and (3) quarterly bonuses help to improve performance of the weak performers by serving as pacers to keep the sales force on track to achieve their annual sales quotas. salespeople. We also find clear evidence of hyperbolic discounting by
1. Introduction Personal selling is one of the most important elements of the marketing mix, especially in the context of B2B firms. An estimated 20 million people work as salespersons in the United States (Albers et al. 2008). Sales force costs average about 10% of sales revenues and as much as 40% of sales revenues for certain industries (Albers et al. forthcoming). In the aggregate, U.S. firms spent over $800 billion on sales forces in 2006, a number that is three times larger than advertising spending (Zoltners, Sinha and Lorimer 2008). Marketing researchers routinely create response models for marketing mix instruments such as price, sales promotion and advertising. Metaanalysis of various research studies estimate that the sales force expenditure elasticity is about 0.35 (Albers et al., 2008) , relative to about 0.22 for advertising (Assmus, Farley and Lehmann 1983) and about 2.62 for price (Bijmolt et al. 2005). While relative sales force expenditure elasticity is useful in determining the relative effectiveness of different instruments in the marketing mix, they give us little insight on how to design a sales force compensation plan, which is widely understood to be the primary tool by which firms can induce the sales force to exert the optimal levels of effort and thus to optimize the use of sales force expenditures. A compensation plan can consist of many components: salary, commissions, and bonuses on achieving a certain threshold of performance called quotas. Figure 1 shows a variety of compensation plans that include combinations of these components. According to Joseph and Kalwani (1998), only about 25% of firms use a pure commissionbased plan; the rest used some form of quotas. As per the Incentive Practices Research Study (2008) by ZS Associates, 73%, 85% and 89% in the pharma/biotech, medical devices and high tech industries respectively uses quota based compensation. This paper has two substantive goals: First, to gain insight on how a firm should design its compensation plan. addition to commissions? Specifically, should a firm offer quotas and bonuses in Despite the ubiquity of quotabased compensation, there is
considerable controversy in the theoretical (e.g., Holmstrom and Milgrom 1987; Lal and Srinivasan 1993) and empirical literature (Oyer 1998; Steenburgh 2008) about the effectiveness of quotas and bonuses relative to straight linear commission plans. Our paper sheds light on this controversy by estimating a dynamic structural model of how the sales
1
force responds to alternative compensation instruments and specific levels of commission rates, quotas and bonus levels. Second, what should be the frequency of bonuses? quarterly or annual bonus? Should one use a monthly,
Should one use a quarterly bonus in addition to an annual bonus?
In the education literature, researchers have argued that frequent testing leads to better performance outcomes (BangertDrowns et al. 1991). Can quarterly quotas serve a similar role to improve outcomes? Like in the education literature, where frequent exams keep students prepared for the comprehensive final exam; frequent quotabonus plans may serve as a mechanism to keep the sales force motivated to perform in the shortrun well enough to be in striking distance of the overall annual performance quota. Methodologically, the paper offers two key innovations. First, we empirically
implement unobserved heterogeneity in a latent class framework within a computationally light two step conditional choice probability (CCP) framework to estimate the dynamic structural model. Though the use of two step estimation approaches have recently gained popularity (Hotz and Miller 1993; Bajari, Benkard and Levin 2007), due to ease of computation relative to traditional nested fixed point estimation approaches (e.g., Rust 1987), their use in empirical applications have been limited by their inability to accommodate unobserved heterogeneity. Arcidiacono and Miller (2010) propose an approach that allows accommodation of latent class heterogeneity within the twostep estimation framework. However, there are few empirical applications of this approach. To the best of our knowledge, ours is among the first empirical papers applying the Arcidiacono and Miller approach to account for unobserved heterogeneity in the twostep dynamic structural estimation framework.2 Second, and of great importance to the dynamic structural modeling literature, we estimate rather than assume discount factors. It is wellknown in the literature on dynamic structural models that discount factors cannot be identified in standard applications because there are no instruments that provide exclusion restrictions across current and future period payoffs (Rust 1994). Hence the standard approach is to assume discount factors. In contrast to this, in our application, we have natural instruments in the form of bonuses: in
Two concurrent working papers that have implemented this approach in economics are Finger (2008) and Beauchamp (2010). 2
2
Hence we need to estimate two discount parameters: a shortrun present bias factor (β<1) and a longterm discount factor (δ). Sudhir (2001) makes an inference about manufacturer actions (wholesale prices) from the observed retail price and sales to infer competition between manufacturers. one observes the agent's action in response to the firm's action. 3 3 . researchers have shown evidence of ‘hyperbolic discounting’ (Thaler 1981). Elster (1979). the presence of bonuses at different frequencies (quarterly and annual) provides us instruments necessary to identify both the discount parameters. one does not observe the actions of the sales force. i. 1998). Laibson (1997.. i. the exerted effort.nonbonus periods. One only observes the outcome of the agent's effort. An individual may prefer 100$ today to 120$ in a year but may also prefer 120$ in two years to 100$ in a year. Hence one has to make an inference about the agent's action (effort) that leads to sales from the observed realized sales. For example. where in contrast to the constant exponential discounting (Samuelson 1937). Fortunately.e. This enables us to estimate discount factors from the data. For example in a consumer response model. the model reduces to the single parameter exponential discounting model.3 The issue has parallels in empirical channel response models. This requires some modeling assumptions on the link between sales and effort. advertising or sales promotion. Hyperbolic discounting is typically mathematically represented using the following quasihyperbolic discount function at time t: D(t)=βδt (Phelps and Pollak (1968). In contrast. which is correlated (but not a onetoone mapping) with effort. When β=1. in a typical structural model. especially those with quotas and bonuses. Further. but only on future payoffs. one observes consumers’ choices in response to the firm's choice of marketing mix such as price. bonuses should have no impact on current period payoffs. sales. First. The key idea of hyperbolic discounting is that individuals discount the immediate future from the present more than they do for the same time interval starting at a future date and hence a declining discount rate.. the psychology literature has shown strong evidence of hyperbolic discounting or present bias. The most frequently given example is the preference reversal shown between two delayed rewards. for a sales force response model.e. There are three specific modeling and estimation challenges in the structural estimation of response to compensation plans.
but the future payoff that she can obtain by being in a more favorable state that can facilitate in obtaining a bonus. This requires a compensation structure of the focal firm features almost all dimensions in typically used compensation plans. when she has a moderate chance of achieving this quota. In addition. but may stretch herself to reach the quota. the As the quotas and bonuses are provided at two different frequencies: quarterly and annual. This implies that the optimal level of effort (and therefore sales) would vary from period to period as a function of the agent's state (distance to quota). a sales agent's optimal effort depends on her state: how close the person is to achieving her quota. The insight is that in any given period. force response behavior. We estimate the dynamic structural model of sales force response to various features of the compensation plan using sales force output and compensation data from a Fortune 500 firm that sells office durable goods. A third issue follows from the discussion of the second. A sales agent may find it optimal to reduce effort when she is close or very far from achieving quota. which would otherwise be impossible. dynamic structural model. it also induces intertemporal dynamics in optimal sales An agent has to be concerned not just with the current payoff by expending effort. This rich plan provides us two key benefits: First. when there is no variability in the plan? Here we draw on an empirical insight from Steenburgh (2008) that can help identify the sales force response. This firm used Plan F in figure 1b. the compensation plan remains stationary over at least a year. since the bonuses are at different frequencies. even though theoretically one can perform counterfactuals of any type of compensation plan if we can estimate structural parameters (other than discount factors) for a sales person with a less rich compensation plan. we observe how the sales force responds to these different dimensions of the plan. we are able to estimate discount factors (present bias factor and longterm discount factor) in a hyperbolic discounting model.A second challenge is that unlike marketing mix variables that change over time. While quotas enable identification of sales force response. Second. How can one identify how the sales force will respond to a compensation plan. when the compensation plan involves payments for reaching quotas. the presence of bonuses allows us to identify and estimate discount factors. This implies that the estimated structural model needs to account for forwardlooking behavior on the part of sales agents. an analyst or manager should have greater 4 . Further.
s=e.faith in the counterfactuals. (2) how a person’s distance to quota can induce heterogeneity in effort. where bonuses occur every few periods. 5 . Consider the pure commission case with no bonus where d=1. r=10 and B=0. with maximum effort of e* when S=5. We conclude the introduction with a numerical example that clarifies (1) how bonuses can efficiently stretch sales people to exert more effort for a given level of compensation. Figure 2a illustrates these results graphically. the commission rate r has to be increased to 20 and would cost the firm $200. relative to when S=3 or S=7. For an intertemporal model. r=10. For the bonus case. i. To achieve the same level of effort from a pure commission plan.. We begin with the case where all agents that are identical in their distance to quota in the current period. S=5 (moderately close to quota) and S=7 (close to quota). In this case. Figure 2b shows the results graphically. But this also means that the sales person needs to look forward dynamically when exerting effort in earlier periods in order to be in a "good" state to reach the quota and receive a bonus in the bonus periods. and B=30. For simplicity. the optimal effort is higher at e*=10.e. S de rs B·1 S Q where –d is the disutility parameter and r is the commission rate (d>0. the decision to stretch to obtain a bonus would depend on how close to bonus the sales person would be in the bonus period. who has sold S units at the beginning of the new period be: U s. Thus the quotabonus plan is more efficient in obtaining the same level of sales. Next let us consider the scenario where d=2. based on parameters that were estimated from observed responses to different elements of the compensation plan. specifically where S=0. e. r>0) and B is the bonus obtained upon reaching quota (Q). The compensation cost to the firm is $130. assume Q=10. Q=10 and B=30. where agents are different in their distance to quotas such as S=0 (far away from quota). Let the utility function of the salesperson that trades off effort (e) and income from sales (s). i.. assume a onetoonemapping between sales and effort. In this case.e. the optimal effort is e*=5. agents have sold nothing thus far.
Related Literature The literature review is in two parts: We begin with the discussion of the literature relating to the substantive issue of the role of nonlinearities such as quotas and bonuses in compensation plans. Using the specific assumptions of linear exponential utility and normal errors (LEN) they show that a linear commission incentive scheme can achieve the best possible outcomes for the firm. literature. Following this. Section 7 concludes. plans offer high powered incentives that can motivate salespeople to work harder (e. Section 2 discusses the related Section 3 summarizes the institutional details of the firm and data used for the empirical analysis. we discuss the empirical literature on structural estimation of worker productivity. Oyer (2000) shows that when participation constraints are not binding. Darmon 1997). quotabonus plan is just one of many possible plans that lead to first best outcomes. 6 . We present the model and the estimation methodology in sections 4 and 5. Park (1995) and Kim (1997) demonstrate that a quotabonus plan may lead to the firstbest outcome. but in their framework. the optimality of nonlinear compensation plans. Rao (1990) also shows a similar result on However. they provide the best compromise between efficiency and ease of implementation.The rest of the paper is organized as follows. Holmstrom and Milgrom (1987) and Lal and Srinivasan (1993) question the need for nonlinear compensation schemes. (1985) apply the principal agent framework of Holmstrom (1979) and demonstrate that a combination of salary plus commission (usually nonlinear with respect to sales) will be optimal. Yet. 2. a quotabonus plan with linear commissions beyond quotas can be uniquely optimal because it can concentrate the compensation in the region of effort where the marginal revenue from effort minus the cost of compensation is maximized. Basu et al.. why do most firms have quota based compensation plans? compensation plans nonlinear? Why are Raju and Srinivasan (1996) suggest that even though a Others argue that quota based commission over quota plan may not be theoretically optimal.g. Section 6 discusses the estimation results and the counterfactual analysis. In the theoretical literature.
however our paper innovates on two key First. Our work is related to several static structural models of worker behavior such as Ferrall and Shearer (1999) and Paarsch and Shearer (2000). However. because we do not have data on selection across contracts. unlike sales force productivity. Much of the variation in productivity here can be explained by observed characteristics such as machine breakdowns etc.There is limited empirical work addressing this issue. there is limited unobserved uncertainty in check sorting productivity. Copeland and Monnet (2008) estimate a dynamic structural model of worker productivity in a checksorting environment with nonlinear incentives. we contribute to the broader dynamic structural modeling 4 This approach to accommodate heterogeneity is similar to the estimation of individual level utility functions in conjoint analysis or scanner panel data.4 Second. he finds that the net improvement in revenues from effort dominates the inefficiencies induced by intertemporal dynamic considerations. A contemporaneous paper by Misra and Nair (2009) on sales force compensation is closest to this paper in methods and substantive context. Oyer (1998) concludes that the negative effects of quota based plans encouraging sales people to maneuver the timing of orders are greater than the benefits obtained from more effort. given linear contracts. Using individual sales performance data from the same firm used in this study (utilizes compensation plan F in figure 1b). Steenburgh (2008) questions whether aggregate data can be used to reach this conclusion. which require us to model the dynamic response of sales agents. who assume discount factors. Further. the approach requires that sales people will exert effort equally across all customersan assumption they show is valid in their data. estimating each sales person’s utility function separately. who endogenize the optimal contract choice of the firm. In terms of methodology. unlike Misra and Nair. 7 . dimensions. we do not model the contract choice. by using a large number of observations related to a particular individual. we seek to understand the response to nonlinear incentives. Our method of using latent classes works in the more common situation where there are limited observations per individual. Based on an analysis of aggregate sales across different industries in different quarters. but unlikely to hold in general. In contrast to these papers. we accommodate latent class heterogeneity within the twostep Misra and Nair sidestep the unobserved heterogeneity issue by estimation framework—an issue that has been an econometric challenge for the literature for close to two decades. both papers use the twostep estimation technique.
This helps us estimate discount factors. the company offers extra overachievement commissions for exceeding quotas and use a group quota updating procedure that minimizes ratcheting effects. Misra and Nair conclude that quotas reduce performance. 3. These “rep” firms do not compete with the direct sales force. Institutional Details and ModelFree Evidence of Dynamics We begin by describing the details of the compensation plan. and as a result this helps us estimate both the present bias factor and the longterm discount factor in a hyperbolic discounting model. Then we provide modelfree evidence to highlight the salient characteristics of the data to be modeled. 8 . therefore. the company followed an explicit policy of ratcheting quotas based on past productivity. which reduced incentives of sales people to work hard in any given period. The Compensation Plan The focal firm under study is a highly regarded multinational Fortune 500 company that sells durable office products primarily using its own direct sales force. the quota ceiling (beyond which sales people receive zero additional compensation) limits the effort of the most productive sales people who would normally have exceeded the ceiling.literature by estimating discount factors (specifically hyperbolic discount factors) using field data. while Misra and Nair consider the role of quotas with floors and ceilings on commissions (plan D in figure 1b). Second. only one Thus the two papers offer complementary perspectives that enhance our understanding about how quotas impact 5 Such rep firms are the focus of Jiang and Palmatier (2010). In contrast.5 Further. The key substantive difference is that we focus on the value of quotas with bonuses (plan F in figure 1b). This is because of two characteristics of their quotas: First. we find that quotas coupled with bonuses enhance performance. bonuses are at two different temporal distances. 3. Second. Two characteristics of the data facilitate identification of discount factors. quarterly and annual. In the plan we consider. bonus payoffs are in the future and are excluded from current payoffs of the nonbonus months.1. First. because it makes future rewards more difficult to attain due to ratcheted higher quotas. performance. each sales agent is given “exclusive” territories. It also uses an indirect sales force through “rep” firms.
we test for the possibility of sales substitution across quarters by sales agents. in terms of performance evaluation within the firm. i. 3. end of the fourth quarter) an annual lumpsum bonus is paid if the annual quota is met and an overachievement commission is given for excess revenues beyond the annual quota. the salary component will be roughly 30% of total compensation. 9 . The firm’s compensation structure follows the pattern in Plan F of Figure 1b and the details of the compensation schedule for the period of analysis are Every month.2 Model Free Analysis We consider three features of the data that can inform model development.. In the first three quarters.1 Seasonality The descriptive statistics of the data is in Table 2. Overall. First. And at the end of the year (i. 3.e. though these are not used for compensation. is the annual quota. Our analysis focuses on sales performance data from 348 sales people during the three year period 19992001. described in Table 1. this suggests that sales agents are highly responsive to quotas and bonuses. Sales is relatively high at the end of first three quarters (March.. for which an agent could obtain commissions or overachievement commissions.sales agent receives credit for a unit of sales and the firm traditionally does not encourage group work or team cooperation among the sales force. we look at the evidence of forward looking behavior induced by bonuses and hence the need to develop a dynamic model. June. for a salesperson that meets all quotas. sales people receive a fixed monthly salary and There are no caps on revenues commissions per volume of sales generated in that month. the company allocates monthly quotas to its sales force.2. The single most important element. September) and much higher in At first glance.e. Figure 3a shows the average revenue generated for each calendar month by the sales force in the study. a quarterly lumpsum bonus is given when each of the quarterly quotas are met. consider the nature of seasonality in the data. Finally. In building its annual and quarterly quota. we Second. the last quarter (December). the firm views a salesperson as having a successful year if the annual quota is met.
We therefore use this variable to account for demand seasonality. First. we see that over 66% of the firms have a December fiscal yearend. For March. they can serve as a proxy for seasonality in demand. we do not use the fiscal yearend information in our analysis. the x axis is the percentage of quarterly quota completed (%QQ).2: Forward Looking Behavior induced by quarterly and annual bonuses We begin by showing scatter plots and the best fitting nonparametric smoothed polynomial (and its 95% confidence interval) of sales revenues normalized by monthly allocated quotas in the quarterly bonus months (March. Why should we expect sales to be so seasonal and coincide by quarter? Given that the focal firm’s products are a B2B business selling discretionary good whose timing of purchase are also flexible. but the peaks are not as large as for Given that we have more direct proxies for seasonality in the form of indirect sales force revenues. Second. purely compensated on commissions. while for December. Figure 3c shows the distribution of fiscal yearends across months for the year 2000 (the midpoint of our sample). The vertical line shows the %QQ and %AQ at which the salespeople on average achieve their monthly allocated quotas. rather than just due to bonuses? How could we control for the effects of Since these salespeople were seasonality? We were able to obtain aggregate sales revenue data from the indirect sales force who is compensated purely on a commission basis. June. 3. Indeed. we can detect . one possibility is that spending may expand during the month in which the fiscal year ends for its customers due to accounting procedures employed by clients. across the board one does not This see a massive reduction in effort when salespeople get closer to achieving their quota. June and September. Two key elements stand out from the graphs. Figure 3b shows the monthly revenue of the indirect sales force. The shape of the monthly revenue for the indirect sales force shows distinct seasonality suggesting demand fluctuates from period to period independent of compensation. End of quarter months also have peaks relative to other months. September. which might explain a significant boost in sales during that month. the x axis is the percentage of annual quota completed (%AQ).2.But could the higher sales at the end of quarters be due to higher demand during these periods. 10 must be partly because of the overachievement commission rate. December. December) against percentage of quota attained by the previous month in figure 4a.
the quarterly bonus in the future does have an impact on behavior. the salesperson on average sells above the monthly allocated quota. Hence distance to annual quotas can serve as an instrument for estimating discount factors. This also suggests an instrument for estimating discount factors. Should the large annual bonus be split into a quarterly bonus (as in other months) and an annual bonus? This can prevent sales people from giving up in November.evidence of forward looking behavior. Early in the year. such chances becomes less likely and sales people respond by reducing their effort. This preliminary evidence also suggests a natural question for managers. even if they are below targets. However. Annual bonuses should have little impact on current payoffs in March. but only on future payoffs. shows the same relationship in the prebonus months (February. because one has a very limited chance of making up the large gap in just two months. provides an opportunity to identify the two discount factors in hyperbolic discounting: a longterm discount factor and a present bias factor. when seen in tandem with the fact that sales agents in similar states in September or December do not seek to accomplish their targets. However. while later in the year. does the salesperson sell above the monthly allocated quota. the differential way in which the sales person responds to temporally different bonuses at any given point in time. salespeople achieve their monthly targets. only at a very high level of %AQ. they achieve their monthly targets only at fairly high levels of %QQ or %AQ. they still have hopes of receiving the large annual bonus by working hard. This is clear evidence of forward looking behavior. Here a new characteristic stands out. the sales person puts in effort to achieve quota. However. even if they do not have a chance 11 . again indicating forward looking behavior. at all levels of %QQ. However. February. later in the year. even though it does not have an immediate impact on the current payoff. In the early months. This is because hard work (and some luck in the form of positive shocks) may give a reasonable chance of attaining the smaller quarterly targets. even at low levels of %QQ. The next set of graphs presented in Figure 4b. one can infer that the sales person is forward looking. In March and June. On its own. even when there is little chance of achieving the quarterly quota. Early in the year (March and June). June or September. May. This suggests again that in the prequarterly bonus months. May and even August. in November. this might simply mean that salespeople are trying to obtain commissions. August and November).
we build a dynamic model of sales force response to the quotabased compensation scheme. 3. firm chooses the annual compensation plan. The timing of the model is as follows: 1.3. Each month. Further. To test this. Specifically to see if there might be any borrowing effects from the last month of a bonus period to the first period of the next bonus period we include a separate variable for the first month of each quarter. We also include individual level fixed effects. Sales substitution across months One possibility is that sales people giving up at the end of the quarter may be doing so to increase the odds of hitting quotas in subsequent quarters by simply not booking the sales in the current quarter. then one should see a negative linkage between sales in month t and month t+1. the shock plus agent's effort determines the agent's realized sales for the period. The first month of quarter also does not have any significant impact even if we separated the effect for people who are “way off target” in the last month and have therefore the greatest incentives to postpone purchases. 4. At the beginning of each year. we do not find any significant coefficient for the first month of quarter (as in Model 2). agents observe their current state and exert effort in a dynamically optimal manner. How these two issues tradeoff is an empirical question. The results were robust and did not vary with alternative definitions of “way off target”. An idiosyncratic sales shock is realized. we report the results of a regression where sales in month t is regressed against sales in month t1 in table 3. Agent receives compensation. which we subsequently address in the counterfactual analysis. We defined “way off target” as those whose previous quarter sales were less than 50% of their quota. At the same time. suggesting little substitution across quarters. If this were true. as seen in Model 3. 12 . 2. 3. Model Based on the modelfree evidence. We therefore do not model substitution across quarters.of reaching the annual quota. and especially between the last month of a quarter and first month of the next quarter. We do not find any evidence of substitution across months. the cost of such a quarterly quota would be that early in the year. the incentive to stretch after reaching quarterly quotas would be reduced.2.
μi . ε it and ωit are the sales and productivity shocks respectively. Steps 23 are repeated each month until the end of the year and Steps 13 are repeated over the years. Sales Response Model Estimating a sales response model with respect to effort poses a challenge because effort is not observed.ϒ ) + ωit D E Here zit are sales shifters. we need to rule out potential endogeneity effects and productivity shocks in the application. it is necessary to assume that the h function is monotonic in zit and ν it . consider a hypothetical scenario.4. This implies that typical endogeneity issues in choice of marketing mix such as price and advertising. we model the sales revenue function (S ) for salesperson i at time t in two parts: (1) a base level of sales independent of effort.ν it .. To appreciate this. even when effort is observed. This requires us to assume away ν it and ωit . We now describe the model in five parts: (i) the compensation plan (ii) the sales agent’s utility function (iii) the state transitions (iv) effort as a function of state variables and (v) the optimal effort choice by the sales agent. Since ν it is not observed to E the researchers. The realized sales of the current period affect the agent's state of the next period. ν it are unobserved shocks to the researchers that can be observed by the sales agents. Then we would estimate the following two equation model one for sales (S). where one can observe effort. one needs to solve for the effort function only on researcher observable characteristics. h is a function (that solves a static or dynamic model) that E maps the consumer observable shifters ( zit and ν it ) to effort. But when effort is unobserved. and one for effort (e) as follows: D Sit = β eit +γ z it + ν it +ε it E eit = h(z it . Give the above discussion. 4. 1. where the marketing mix choice is correlated with certain market characteristics that are observable to the decision maker. and zit are effort shifters.e. i. parameterized by 13 . cannot be accounted for. but unobservable to the researcher.
and end of year bonus Biyt for achieving the corresponding annual quota Qiyt (3) commission rate rit per dollar worth of sales and an overachievement commission rate. (ii) endof quarter bonus. we use the following variables: Given that effort is a E function of demand shifters. we allow for salespeople to belong to one of multiple discrete segments.D demand shifters ( z it ) and (2) sales induced due to effort (e ). 4. 2 Compensation Plan The compensation plan has three components. we use annual quota from the previous year ( AQ iy −1 ). For the effort shifters in eit. S where ε D f z it e E z it ε … 1 is an additive sales revenue shock that is not anticipated by the salesperson when choosing effort levels. (ISt). tenure with the firm (Tenure) to moderate the level of effort. Specifically. we use the indirect sales revenues for month t. Biqt for achieving the corresponding quarterly quota Qiqt. we allow a timeinvariant salesperson specific variable. They are: (1) the monthly salary wit. To account for the crosssectional variation in market potential. we include both AQ y −1 and ISt in z it . which is a function of effort E shifters that include both territory and salesperson characteristics ( z it ). As discussed in the motivation. but also has a time varying seasonal component. We estimate the effort function nonparametrically. hence these effort functions will be estimated at the segment level. %AQit) as variables that affect effort. Note that unlike the demand shifter function f. As discussed earlier. the market potential varies across territories. In addition. To account for seasonality of demand across months. We therefore use the cumulative percentage of quarterly and annual quota completed till time t (%QQit. We also include an interaction between the two variables to account for the possibility that seasonality will have a larger impact on larger territories. rit′ given at the end of the year 14 . the effort function will vary across salespeople. which is common across all salespeople. the salesperson’s state with respect to achieving quota will have an impact on the effort they expend. by using Chebyshev polynomials of the variables described above.
rit′}. and the overachievement commission component OCit. In the case of the CARA utility function (exponential utility function) with normal errors and a linear compensation plan. 4. 3 Sales person’s perperiod utility In each period t. θ . α εt Q t where zi1t and zi2t are the percentage of quarterly and annual quotas completed respectively by salesperson i until time t. D . γ W S . α E ε εt . S t . sales person i receives positive utility of wealth Wit earned based on realized sales and a disutility C(eit. and the compensation plan. Thus the utility function is defined as: U e . θ 6 where γi and θi are each the risk aversion and disutility parameters respectively for salesperson i. θi) from exerting effort eit. Q t εt · Bq 1 ·1 t E . Bt Ct OC t E . Wt Bt wt 1 ·1 t The detailed expressions of wealth is as follows. Biqt. Here we consider the utility function to be a second order approximation to a general concave utility function. Qiqt. ψ C e . the commission component Cit. We represent the compensation plan for a salesperson i by the vector ψit ={wit. Q t · r′ t · z t ·Q t s e z E . Biyt .for sales over and above the annual quota for each individual i at time t. the lumpsum bonus component Bit. ψ γ var W S . Wit can be computed. Qiyt. ψ . Wit arises from four components. D . z D . this functional form represents the certainty equivalent utility of the agent. 15 6 . rit. Given the sales levels. z D . the wealth for individual i. the per period salary component wit. 1qt and 1yt are indicators for whether time t is a quarterly or annual bonus period. Q ·B Ct OC t r · s e 1 ·1 t z E . D .
of which we use previous year’s annual quota. the percentage of quarterly quota completed. We also observe individual characteristic. 1. 4 State Variables As discussed. z . the percentage of annual quota completed. 16 z . if t is start of annual quota period S . specifically tenure with the focal firm (τ). and the state variables and their transitions. other wise Q z 2.In our empirical analysis. the nonlinearity of the compensation scheme with quotas and bonuses introduces dynamics into the sales agent's behavior because there is an additional tradeoff between the disutility of effort today and a higher probability of lumpsum bonus and overachievement commissions tomorrow.τ . C e. the salesperson may choose to leave the firm. Percentage of annual quota completed (%QQ) 0. if t is start of quarterly quota period S z . each sales agent would choose an effort level conditional on her states to maximize the discounted stream of expected future utility flow. indirect sales. Naturally.5 Optimal Choice of Effort Given the parameters of the compensation scheme ψ. and territory characteristics. other wise z Q z Other state variables would include time varying demand shifter. Percentage of quarterly quota completed (%AQ) 0. Thus the set of structural parameters of the θ . specifically. the time varying indirect sales is a onetoone mapping to period type and hence includes the information about different periods. .γ . These state variables are collected in a state vector z E 4. IS . AQ . and therefore use it as an individual state variable that impacts effort. θ θ . we use a quadratic functional form for the disutility function. salesperson that needs to be estimated are Ω 4. To incorporate the dynamics of the model we These state variables evolve as follows: consider the following stochastic state variables. if this value function is below the reservation wage. Alternatively.
Ω where Ω is the primitives or the structural parameters of the underlying utility function. ψ. Ω or differently put as. Ω V z′. and Levin (2007) can serve to reduce the In this approach. parameter space for every iteration. The expectation of the value function is taken with respect to both the present and future sales shocks. β and δ are the discount parameters. the conditional choice probabilities of choosing a certain action as a function of state variables are estimated in a flexible nonparametric manner. in the second step. Ω e βδ max U e. Ω max U e. ψ. ψ. the model reduces to a single parameter exponential discount model. conditional on staying at the firm. NFXP estimators are computationally The twostep estimation first introduced by Hotz and burdensome as one has to solve the dynamic program numerically over each guess of the Miller (1993) and extended by Bajari. For this approach to work. Ω e βδ V z .The stream of utility flow. Ω δ V z′′. ψ. ψ. 5. under the optimal effort policy function. The longterm discount factor is δ. If β=1. V z. Ω max U e. V z. Until recently. ψ. 17 . z . ψ. In the first step. Benkard. However. Ω δ max U e. Then. Ω max U e. specifically the disutility parameter θ and the risk aversion parameter γ. it is critical that the conditional choice probabilities are estimated accurately in the first step. z′. ψ. where β<1 represents the present bias in hyperbolic discounting. ψ. Estimation Traditionally. the shortterm discount factor is βδ. computation burden. these conditional choice probabilities are used to estimate the structural parameters of the sales agent's utility function. the model estimation proceeds in two steps. z. and the behavioral notion of quasihyperbolic discounting can be represented by a value function. the nested fixedpoint algorithm (NFXP) developed by Rust (1987) is used to estimate dynamic models. z. z′′. it was believed that the accurate estimation of conditional choice probabilities for an agent is impractical when there is unobserved heterogeneity. ψ.
which is a subset of zitE (from now on refer to as zit). 18 . come from a normal distribution that are i. The revenues from the indirect sales force capture market seasonality.. estimation procedure. Thus the nonparametric effort function is: ρℓ z E · λ ℓ … 2 ℓ We now discuss the details of the two step We model the effort function nonparametrically as a combination of basis functions of the state variables. (2) the revenues of the indirect sales force. across time. order Chebyshev polynomial. 5.e. the ℓth basis function is the ℓth From equations (1) – (2) we have the following sales response function to estimate. compensation plan.i. The interactions effects of these variables with the other state variables goes into the polynomial function in (2).Arcidiacono and Miller (2010) have proposed an EM–Algorithm based approach to accommodate unobserved heterogeneity in the first step of the two step estimation procedure. e Where the ℓth basis function is ρℓ z E . market size. we use two variables: (1) lagged annual quota for salesperson i. α ℓ S ρℓ z E · λ ℓ ε For zitD. we need to estimate a flexible nonparametric mapping between observable states and actions of the sales person. that links effort and state in equation (1). i. independent of the nonlinear nature of the We assume that the revenue shocks (ε ). The lagged annual quota takes into account general territory characteristics that are likely to be generated independent of effort. f z D .d. We use the direct linear effect of these variables to control for cross sectional variations of territory characteristics and temporal variations in monthly seasonality. We provide one of the first applications of this approach – illustrating the empirical validity of the approach in practical applications. this requires a nonparametric model of the monthly effort function e z E . 1 Step 1 In this step. In this application.
and the unobservable segment k. The parameter Θ the sales response and effort policy function where each λk is the parameters that index the effort policy for segment k and σk is parameter for the distribution of the revenue shocks for segment k. history zi. given the observable L S  z . where L ε e … (4) √ We accommodate unobserved heterogeneity by allowing for discrete segments. given segment parameters Θk.If one could estimate the sales response and effort response function at the level of each individual. Θ . Θ be the likelihood of individual i's sales being Sit at time t.λ . Θ. k ∈ {1. π . k.K} with segment probabilities qi={qi1. qiK}.…. and the unobservable segment k is given by: T the likelihood of observing sales history Si over the time period t=1…T.σ is the vector of segment level parameters of shocks to be normally distributed and hence use the normal likelihood for equation (5) as in equation (4). By summing over all of the unobserved states k∈{1. As noted earlier we assume the distribution of the revenue α . Let S z . we obtain the overall likelihood of individual i: K L S  z .…. Θ . π q … 5 where S  z .λ . π q L 19 S  z .K}.…. … 3 where the vector Θ α . α ℓ ρℓ z E · λ ℓ . we can simply obtain the individual level parameters of the effort and sales policy function by maximizing the log likelihood of the sample such as T L D Θ argmax log L S f z . Assume that sales person i belongs to one of K segments. Θ . k. Let the population probability of being in segment k be πk.σ contains the set of parameters of the sales response and effort policy functions and the distribution of sales shocks. conditional Then on the observables zit.
and hence the loglikelihood over the N sample of individuals becomes N N K T log L S  z .7 Natural candidate for such starting values would be to obtain the parameters from a model without unobserved Given the parameters {Θm. π S  z . the update of the (m+1)th iteration is as follows (a) Compute qik(m+1) using equation (8) with Θm and πm (b) Obtain Θm by maximizing (7) evaluated at qik(m+1) (c) Update π(m+1) by taking the average over the sample such that π 1 N N q We would iterate (a) – (c) till convergence. Θ. π … 8 L S  z . z . and conditional on all of the observed data of individual i. π q S . Θ . segment probabilities π. Θ. πm} from the mth iteration. We start with an initial guess of the parameters Θ0 and π0. Pr kS . π log q … 6 Directly maximizing the loglikelihood in (6) is computationally infeasible because the function is not additively separable so we take the approach of Arcidiacono and Jones (2003) and Arcidiacono and Miller (2010) to iteratively maximize the expected loglikelihood in equation (7) N K T q log S  z k. The initial values of the segment probabilities were set equally across segments. Θ … 7 where qik is formally defined below as the probability that individual i is of segment type k given parameters values Θ. heterogeneity and slightly perturbing those values. Θ. z . We started the initial values from one tenth of the standard error from the parameter values obtain from a single segment model. 20 7 . Θ. π The iterative process is as follows.
S. conditional on the compensation plan ψ. Using these observed optimal actions we are able to construct estimates of the value function. we use Chebyshev polynomials of state variables to approximate effort. 1998. e. Ω z z. More formally. z . S. Ω … 9 where D t is the hyperbolic discount factor. ψ. ψ.For the basis functions in the effort policy. Kenneth L. Using the estimated sales and effort policy function and the distribution of the sales shocks in the first stage.8 As a result. e . we are able to forward simulate the actions of sales agents to obtain the estimate of the value function. we obtain the sales revenue function S . 8 For reference.λ . Judd. and the expectation operator would be over the present and future sales shock εt. ΘK π α . see “Numerical Methods in Economics”. πK and effort policy function Therefore. 5. other wise D t U e z . Ω 1. probabilities for each segment. The detailed simulation procedure is as follows. we obtain the vector of parameters that index these basis functions λ′s. ε . for each segment. … . MIT Press. and the parameters of the revenue shocks σ’s for each segment k. Let the value function of a representative agent at state z that follows an action profile e. 21 . … . the sales profile S and the primitives of the utility function Ω be represented as T V z. which enables us to estimate the primitives of the model that rationalize these optimal actions. Θ Θ . if t 0 βδ . 2 Step 2 The key idea of the twostep estimation is that in the 1st stage we observe the agent’s optimal actions.σ Also we obtain the population segment π . the vector of parameters for the sales policy α′s.
ψ.e.Ω) would always have value of greater or equal to zero. 22 10 9 . S. e z . S.11 For each segment. ψ. As indicated in Bajari. ψ. e z . ψ. Q v. S. ψ. Ω argmin min Q v. by using the same simulation method proposed above. ψ. Our empirical counterpart to V z. S. Ω result our estimates of the structural parameters are obtained from minimizing the objective function in equation (10). the function Q(v. Ω . S. S. ψ. S. Ω V z.10 policy. S. Ω . Although the method of selecting a particular perturbation will have implications for efficiency the only requirement necessary for consistency is that the distribution of these perturbations has sufficient support to yield identification. ψ. i. e z . Ω Then if e(x) is the optimal Thus denotes a particular {z. e z . our estimate of the underlying structural parameters Ω would satisfy. let the corresponding estimate of the value function be called the suboptimal value function V z.9 Let es(z) be any deviation policy from a set of feasible policies that is not identical to the optimal policy and. Ω and as a V z. Ω where v V z. es(z)} combination. Ω would be Q v.ψ.(a) From initial state of zt calculate the optimal actions as e(zt) (b) Draw sales shock εt from f(ε) (c) Update state zt+1 using the realized sales s(e(zt))+ εt (d) Repeat (a) – (c) until t=T By averaging the sum of the discounted stream of utility flow over multiple simulated paths we can get the estimate of the value function V z. Since e(z) by definition is the effort policy and thus at an optimum. We chose to draw a deviation policy from a normal distribution with mean zero and quarter of the variance from the revenue shock distribution. S. we drew four hundred simulation draws over each period and computed the value functions. ψ. there are multiple ways to draw these suboptimal policy rules. es(z)=e(z)+η 11 We drew two hundred deviation strategies to construct the objective function and hence NI=200. e z . S. then any deviations from this policy rule would generate value functions of less or equal value to that of the optimal level. e z . Ω . Let us define the difference in the two value functions as. and Levin (2007). Ω of inequalities. ψ. 0 2 dH v where H(v) is the distribution over the set Q v. Benkard.
0). However. Ω . min(V z. a sales person can choose to increase effort to increase the likelihood Given multiple observations of sales at different states. S. This allows us to estimate the structural parameters for each segment. The choice of effort in nonbonus periods has two distinguishable wealth effects. Further. In our model.12 5. S. Thus variation in sales (which is monotonically linked to effort) as a function of these state over time helps identify the effort disutility parameter. a sales person would just simply solve the myopic first order condition every period to decide on the optimal level of effort. the state variables (%QQ and %AQ) serve as identifying instruments. e z . The constraint was nonbinding and did not impact our estimates. The risk parameter is identified by differences in response to variations in wealth levels over time. Identification We provide a brief and informal discussion of identification.e. relationship nonparametrically. ψ.1 NI NI min Q v . a deterministic function of effort.3. Without these future bonuses. The discount factor is typically not identified in dynamic choice models because instruments affect both the current utility and future utility (Rust 1994. i.. 23 12 . Magnac and Thesmar 2002). (1) the increase in current utility through the commissions and (2) the increase in future utility through bonuses. Ω . 0 … 10 The above procedure is performed for each segment with the segment specific effort policies obtained in Step1. with the future bonuses. we assume a deterministic (but highly flexible) relationship between effort and observable states (percentage of quarterly and annual quotas achieved by previous month and the demand shifters) at the level of each segment. we can separately identify nonparametrically the density of sales shocks and We make a parametric assumption about a strictly monotonic relationship between sales and effort because it is not possible to identify this We also estimated a model with a second set of moment inequalities to reflect the participation constraint that employees continued to work at a firm because they at least obtained a reservation value (normalized to zero). ψ. Realized sales is a function of effort and additive sales shocks.
Uniformly. conditional on the state variables and depending on how much she values the future. which by definition is the discount factor. 6. having two different future bonuses. 6. We estimate segment level effort policy functions by estimating the nonparametric relationship between sales and state variables through Chebyshev polynomials of the state variables. one in the near future (quarterly bonuses) and the other in the distant future (annual bonuses and overachievement commissions) with different state variables affecting them. In the second stage estimation. Further. then we report estimates of structural parameters of sales agents' utility functions from the second stage estimation. We estimated up to fourth order Chebyshev polynomials with alternative number of segments and choose the best fitting model based on the Bayesian Information Criterion (BIC). As the coefficients associated with the Chebyshev polynomials have no intuitive meaning. Results We first report the first stage estimates of the demand shifters and effort policy function for the sales response model. We then perform several counterfactual simulations to address the substantive questions we seek to answer. Thus larger markets tend to have a bigger sales multiplier independent of effort in high demand periods. we show graphs of the effort policy function for the three segments as a function of percentage annual quota (%AQ) for March (end of first quarter) and December (end of year) 24 . for intuition. we perform a grid search over the discount parameters (betas and deltas). the modelfree evidence in Figure 4 shows different responses to annual and quarterly bonuses. 1 First Stage Estimates The parameter estimates for the demand shifters in the sales response function is reported in Table 4a. The estimates of the best fitting Chebyshev polynomial function (ρn indicates the nth order Chebyshev polynomial) and the standard deviations of the revenue shocks for each segment are reported in Table 4b and 4c. Indeed. We find that only the interaction term between lagged annual quota and indirect sales revenue are statistically significant. help us identify the hyperbolic discount parameters. three segment models had the best fit.of bonus.
the lowest productivity segment.1 indicates 10% above quota. who finds that sales people “give up” when far away from achieving quota. This is probably due to the fact that in the early years of their careers. Segment 2 is the largest with a share of 47%. 6. followed by Segment 2 and Segment 1. Figure 5a shows the Segment 3 exerts the most effort and is the most productive segment. likely due to the overachievement commissions that incentivize sales people to keep exerting effort even after achieving quota. Our results are consistent with Steenburgh (2008). Given that the average states in March for each segment were 0. 0.2 Discount Factor 25 .55. We also see a positive relationship between exerted effort and %AQ for all months shown.62. many sales people are not in a position to slow down. does not gain in productivity from experience. As for %QQ. Figure 5b shows the effect of tenure on effort for all segments. Segment 3 has the highest annual quotas. but this tapers off with time. after a certain number of years in the same job. they want to work hard not only for monetary payments from increased wages but also other intangible incentives such as promotions or transfers to better job titles. This is consistent with the allocated quotas and percentage of time quotas are achieved in Table 3. respectively. but do not slow down much once quota is reached. and Segment 1 exerts the least effort and is the least productive segment. Sales people in segment 2 and 3 initially increase effort with experience.58. However. Table 5 shows the share of the three segments and their descriptive characteristics. such that 1 implies at quota and 0. 0. these career concerns do not matter as much and the effort levels tend to taper off. Segment 1. we see an increasing but concave relationship in March implying that once a sales person is way above the quarterly quota she starts to gradually slow down.9 indicates 10% below quota and 1. Interestingly.in Figure 5. The average tenure with the firm is not very different across segments at approximately 12 years. %AQ is normalized across sales agents. Effort in December does not fall off even if the sales person has already reached or exceed quota (%AQ>1). such as for all segments in our case. Interestingly. Segments 1 and 3 have shares of 32% and 21% respectively. Segments 2 and 3 with larger quotas achieve their quota targets more often than Segment 1 which has trouble meeting quota.
02 to no For purely monetary values. not accounting for the large future bonuses. Table 6 presents the mean absolute percentage errors (MAPE) A beta of 0.7%. In a static model.1 for beta. 6. the disutility parameters for all three segments are negative and significant.8 is optimal. the estimated discount factor seems rather low. This may be because in the range of incomes earned by the sales force. exponential discounting. But as Frederick.8 and a delta of 0. Segment 1. discounting at all with a discount factor of 1. first stage. consistent with the model we developed earlier. Overall. 3 SecondStage Structural Parameter Estimates The first column of Table 7 reports the structural parameter estimates of the sales agent's utility function for a forward looking sales person. which produces the greatest sales on average. This will underestimate the 13 We tested for finer granularity around beta for 0. The summary shows that the estimated discount factors vary extensively ranging from as low as a mere 0. Loewenstein and O’Donoghue (2002) point out. These estimates are consistent with the effort policy functions estimated in the Segment 3. has the lowest disutility for effort. The risk aversion coefficients for all segments are insignificant showing no direct evidence of risk aversion by the sales agents. Loewenstein and O’Donoghue (2002) have a comprehensive summary of the estimated discount factors from previous studies. Frederick. for behavioral aspects such as pain and thus in our case effort. The estimated model fits the observed sales revenue data reasonably well with a MAPE of 10. risk aversion is not a serious concern. any effort would be attributed to current payoff.92 has the lowest MAPE. 6. the discount factors tend to be low and hence our estimates appear reasonable. has the greatest disutility.01 for delta and 0. which has the lowest sales. 4 Assessing the value of a dynamic structural model How important is it to account for dynamics to model salesperson behavior? If salespeople are behaving in a dynamically optimal manner.13 Thus associated with each set of hyperbolic parameters where a beta equals one represents our estimates show a distinct present bias in that β <1.8 and found that 0. not accounting for dynamics would bias the structural parameters.We performed a grid search over the set of discount parameters in steps of 0. 26 .
5 Counter–Factual Simulations We now perform a series of counterfactual simulations that address the two sets of substantive questions we wish to answer. uncertain incomes in the development economics literature. As predicted. given the uncertainty in future demand shocks. We next compare the revenue and effort predictions between the dynamic and myopic models. but set the discount parameters to zero for the myopic model.7% for the revenues are systematically lower for the myopic sales agent because the sales person does not take into account the effect of future bonuses and overachievement commission in current effort. For Segment 3 the bias is as large as 22% relative to the dynamic model estimate. To isolate the effects of forward looking behavior. myopic model also has a poorer fit: a MAPE of 18. 6. so they prepare for such a rainy day by working harder early on so that they are within striking target of quota even if a bad sales shock occurred. The myopic salesperson exerts much more effort in the bonus period. which may prevent getting to the quota. we use the structural disutility parameter estimates from the dynamic model for both the forward looking and myopic models. the disutility The parameters are smaller in magnitude relative to the forward looking model for all segments. The overall change in revenues under 27 The observed effort smoothing is similar to consumption smoothing by forward looking consumers facing . Figure 6 The compares the predicted revenues and effort of the myopic and dynamic models.salesperson disutility parameters and overestimate the effects of compensation on productivity. there is a chance of bad shocks later. we address the issue of how valuable different components of the compensation plan are.8% relative to the MAPE of 10. The effort peak in the bonus periods are not as pronounced for the dynamic consumer. dynamic model. The forward looking sales person anticipates that in an uncertain environment. The second column of Table 7 shows the estimates of the structural model without forward looking dynamics – discount factor set to zero. but the forward looking sales person smoothes effort over time. First. The effort graph in Figure 6 enables us to isolate out the sales revenue cyclicality and focus on the differences in effort across dynamic and myopic agents.
we find from Table 9 all segments suffer from substantially poorer performance when quotas are removed and the firm shifts to a pure commission scheme.the alternative conditions is reported in Table 8 and the effect by segment in Table 9.8% greater with the current compensation plan compared to a pure commission plan. those who did not meet quota decrease effort towards the end of the year as they are unlikely to meet quota and therefore overachievement commission has no impact on their earnings. profits are lower by about 2% (assuming gross margin of 33%). and (2) a higher commission rate is such that total compensation is exactly equal to the current compensation. we find that revenues are about 4% higher. Second. Value of Overachievement Compensation We compare changes in revenues and profits when the firm eliminates the overachievement commission rate. For those who met the annual quota. In contrast. Interestingly. Thus overachievement commission provides the incentives for the most productive sales people even if they have already met quota (or likely to meet quota). which motivates sales people who are close to reaching their quota to continue exerting effort. respectively. We consider two cases: (1) where the commission rate is the same as the current commission rate. the effort level does not decline even when close to the quota because of the overachievement commission. Value of Quotas and Bonuses We compare changes in revenues and profits when the firm moves from the current compensation plan to a pure commissiononly plan.3% and even accounting for the additional commission costs. Not surprisingly Table 9 indicates that overachievement commissions have the most impact on Segments 2 and 3. Figure 7 plots the effort level of sales agents who met and didn’t meet the annual quota. Even when adjusting commission rates to be higher to make total compensation identical to current levels. Overall revenues drop by 13. We find that the revenues are about 20. we assess compare the role of bonus frequency— how quarterly and annual bonuses affect performance. Value of Cumulative Annual Quota 28 .
we remove the overachievement This decrease is greater than Thus the cumulative commission (which is based on reaching the annual quota) and split the total bonus payments Overall. up meeting quota. To study this. In effect. it reduced revenues overall Total revenues drop by 1%. because sales people did not put in as much effort earlier in the year to be within striking distance of annual quota. The impact of quarterly bonuses also differs across the three segments of consumers. revenues drop by 15. Why? Not only does the quarterly quota induce sales agents to work harder in a given quarter but it also helps them achieve the annual quota by giving the incentive to stay on track of their annual goals. the most productive segment. The Role of Quarterly versus Annual Quota We next investigate the value of quarterly bonuses relative to annual bonuses. We also consider the case where we split the annual quota into a quarterly bonus and an annual bonus so that people do not “give up” in the last quarter when they are far away from quota. quarterly bonuses Hence. Overall revenues fall by 5%. because it is not as large. but very high impact on Segment 1.3%. Thus annual quotas and overachievement commissions have less of an impact on yearend performance without quarterly bonuses. But this lack of incentive leads them to be farther away from the annual quota by December.Rather than have a cumulative annual quota. Table 9 indicates that quarterly bonuses have relatively minimal impact on Segment 3. the 13. Effort drops consistently across the year when there are no quarterly quotas. when there is the annual bonus on the table. When the quarterly quota is removed.8%. revenue falls by 2% and effort falls by 4%. annual quotas and overachievement commission have little impact on effort as sales agents are more likely to give 29 . annual quota induces sales agents to exert greater effort and raise revenues by 2. While this did increase the effort in the last quarter. sales agents no longer have as much incentive to work hard early on. Even in December. where we just dropped the overachievement commissions. what would be the effect of replacing it with just a fourth quarter quota? across all four quarters. Figure 8 shows the comparison of effort between the current plan and when quarterly bonuses are eliminated and only an annual bonus is paid at the end of the year.5%.
This paper developed and estimated a dynamic structural model of sales force response to a compensation plan with many components: salary. Our analysis helped us assess the impact of (1) different components of compensation and (2) the differential importance of periodic bonuses on performance on different segments of sales people. In the absence of quarterly bonuses. 30 A key finding . There has been a fair amount of controversy on the value of quotas and bonuses in the literature. quarterly bonuses serve as a continuous evaluation scheme to keep sales agents within striking targets of their annual quotas. commissions. Thus. lumpsum bonus for achieving quotas. i. 1991) and some experimental work in the behavioral psychology (Heath. 7.. a failure in the early periods to accomplish targets caused agents to fall behind more often than in the presence of quarterly bonuses.e. 1999). The basic idea is that achieving shortterm goals make achieving longterm goals more feasible. Larrick and Wu. there has been no analysis todate on the role of bonus frequency in enhancing sales productivity. we find that the quotabonus scheme used by this firm increased performance of the sales force by serving as stretch goals and pushing employees to accomplish targets. There has been some descriptive work in the education literature on how frequent testing affects academic performance (for an extensive survey. Conclusion Even though personal selling is a primary marketing mix tool for most B2B firms to generate sales. Overall. To the best of our knowledge. there is little research on how the compensation plan used to motivate the saleforce affects performance.are needed as pacers to the less productive sales people than for the most productive sales people. in education terms it may imply that weaker students gain more by periodic testing. see BangertDrowns et al. Further. Features such as overachievement compensation reduce the problems associated with sales agents slacking off when they get close to achieving their quota. Our analysis show that the short term goals are more valuable to the least productive segment. relative to stronger students who would study independent of exams. and different commission rates beyond achieving quotas. the quarterly bonus serves as a valuable subgoal which helps the sales force stay on track in achieving their overall goal.
while in other periods. we find that bonuses have an overall positive impact on all segments of salespeople.g. For the most productive segments. 31 . The approach is flexible. Looking at scenarios where contracts varied over time. because demand peaks that coincide with quota periods may be wrongly interpreted as a byproduct of compensation. Features of our data allow us to separate seasonality in sales due to quotas as opposed to underlying consumer demand seasonality. This we believe is an exciting area for future research. identify such a multidimensional effort merely from the sales data as in this paper. while quarterly bonuses are most helpful to increase performance among the weaker performers. Overachievement commissions serve to increase performance among the highest performers. It would be useful to extend structural analysis to settings involving team based compensation. Paarsch and Shearer 1998). One needs more work on scenarios with richer contracts. This enables us to get better estimates of the response to compensation. can also shed Papers that have looked at varying contracts over time typically have focused on only contracts with linear commission rates (e. one can use attrition information to address this issue. This paper does not address the selection issues. By looking at a longer panel of sales people's performance. Finally. effort tends to be multidimensional and one possibility is that quotas and It is not possible to bonuses force people to focus on the effort that lead to final sales in bonus periods. We now discuss limitations of the paper. compensation contracts can serve as a selection mechanism to draw the right type of sales people into the sales force. quarterly quotas does not seem to matter much. Chan. they may focus on earlier stages of the selling process. Nevertheless new data from CRM databases which track customer stages through the selling process can help shed insight on this issue. We use recent innovations in the twostep dynamic structural model estimation to accommodate unobserved heterogeneity in sales force response.. In summary. Li and Pierce (2009) investigate the effects of peer effects on sales performance in the presence of team based compensation in a reduced form analysis. which provide promising avenues for future research. Second.is that annual bonuses are not as effective for the sales force without quarterly bonuses for the low productivity segments. First. light on this problem. yet computationally feasible with minimal additional burden compared to traditional twostep methods.
a reality at many firms. the issues raised analyze this problem and obtains useful substantive insights. How employees respond to such rich compensation structures with bonuses. commissions. has not This paper illustrates a rigorous framework to Nevertheless. quota and bonuses at quarterly and annual frequencies.In summary. this paper provides important insights on how the sales force responds to a very rich compensation structure involving many components of compensation: salary. 32 . been investigated at all in the literature. above provide an interesting agenda for future work.
13311370. Econometrica.References Albers. Srinivasan and Richard Staelin. 109113. Lanier Benkard. Sönke. “Introduction: Special Issue on Enhancing Sales Force Productivity”. Murali K. St.A. 85. Harald J. pp. “CCP Estimation of Dynamic Discrete Choice Models with Unobserved Heterogeneity”. 1984 Bajari. 75(5). Journal of Educational Research. 2005 Chan. Amyia K. Van Heerde and Rik G. Gert. 21. pp. Patrick. 2009 Copeland.. and Shrihari Sridhar. Sonke. 2010 Assmus. Tat Y. Working Paper. 1991 Basu. pp. 8999. pp. Tammo H. Murali K. 2009. Farley and Donald R. 8 (3). Rene. Robert L. Duke University. pp. “Selecting Appropriate Sales Quota Plan Structures and Quotasetting Procedures”. Peter and Robert A. Darmon. James A. Journal of Marketing Research. Journal of Marketing Research 42. Working Paper. 324342. 1997 33 . Journal of Personal Selling and Sales Management. 2010 Bijmolt. “Competition and Peer Effects in Competing Sales Teams”. 76. Jia Li. “Personal Selling Elasticities: A MetaAnalysis”. Kulik. Journal of Marketing Research. Duke University. John U. Washington University. Lehmann. Rajiv Lal. 116. pp. Mantrala. Adam and Cyril Monnet. Miller. 2007 BangertDrowns. V. C. Mantrala.M. 141–56. “Estimating Dynamic Models of Imperfect Competition”. pp. forthcoming Arcidiacono.. Kulik and ChenLin C. Working paper.. 96113. “Salesforce Compensation Plans: An Agency Theoretic Perspective”. “New Empirical Generalizations on the Determinants of Price Elasticity”. 1985 Beauchamp. Review of Economic Studies. “How Advertising Affects Sales: MetaAnalysis of Econometric Results”. “Effects of frequent classroom testing”. 2008 Albers. Pieters. Srinath Gopalakrishna and Kissan Joseph. Journal of Personal Selling and Sales Management. Andrew.. 17. 28. Louis. and Lamar Pierce. Marketing Science. pp. “Abortion Supplier Dynamics”. “The Welfare Effects of Incentive Schemes”. and Jonathan Levin. 6574.
899913. Working Paper. 10. "Limited Liability and Bonus Contracts" Journal of Economics and Management Strategy. “Aggregation and Linearity in the Provision of Intertemporal Incentives”. MIT Press. Son Ku. Larrick and George Wu." Review of Economic Studies. pp. 147159. Joseph.. Duke University. 309338. 1997 34 . Bell Journal of Economics. 79109. Industrial Marketing Management. 1979 Holmstrom. pp. Chip. 1998 Kim. “Time Discounting and Time Preference: A Critical Review”. 1987 Hotz. and Robert A. Review of Economic Studies. V. Kissan and Manohar Kalwani. Journal of Economic Literature. Cognitive Psychology. “Research and Development Competition in the Chemicals Industry”. George Loewenstein and Ted O’Donoghue. 351401. 7491. Bengt and Paul Milgrom. “Incentives and Transactions Costs within the Firm: Estimating an Agency Model Using Payroll Records. “Numerical Methods in Economics”.Elster. “Conditional Choice Probabilities and the Estimation of Dynamic Models”. 1998 Judd. 55. Stephen R. 1999 Finger. 1997 Laibson. “Goals as Reference Points”. 2008 Frederick. pp. Kenneth L. 2002 Heath. Quarterly Journal of Economics. and Bruce Shearer. David. Shane. 61. pp. pp. UC Davis. pp. pp. 443477. Joseph. Miller. 265289. Bengt. Jon. pp. 1999 Holmstrom. 66. “Moral Hazard and Observability”. 38. pp. Palmatier (2009) “Structural Estimation of a Moral Hazard Model: An Application to Business Selling. Christopher. 112.. “The Role of Bonus Pay in Sales force Compensation Plans”. Econometrica. Richard P. 1979 Ferrall. Renna and Robert W. 40. “Golden Eggs and Hyperbolic Discounting”.” Working Paper. 1993 Jiang. 27. 6. “Ulysses and Sirens: Studies in Rationality and Irrationality” Cambridge University Press. 303328.
and Incentive Effects: Statistical Evidence from Payroll Records”. J. “Quota Dynamics and the Intertemporal Allocation of Sales–Force Effort”. and Bruce Shearer. International Economic Review. 14541462. 1995 Phelps. pp. 113. 42. ”Compensation Plans for Singleand Multiproduct Sales forces: An Application of the Holmstrom – Milgrom Model”. Paul. 42. “Incentive Contracting under Limited Liability”. “Identifying dynamic discrete choice models”. S. Econometrica. 2000 Paarsch. 861871. “Fiscal Year Ends and Nonlinear Incentive Contracts: The Effect on Business Seasonality”. pp.. 39. 18. Journal of Economics and Management Strategy. ''Piece Rates. David. 1999 Paarsch. 149185. EunSoo. and Bruce Shearer. Edmund.Laibson. Thierry and David Thesmar. 2002 Misra. Rajiv and V. 405426. 1998 Oyer. “A Theory of Sales Quotas with Limited Liability and Rent Sharing”. Harry. Jagmohan S. 70. European Economic Review. Srinivasan. “On SecondBest National Saving and GameEquilibrium Growth”. Sanjog and Harikesh Nair. 801816. 1998 Lal. pp. Srinivasan. 1996 35 . 1968 Raju. 1993 Magnac. “QuotaBased Compensation Plans for MultiTerritory Heterogeneous SalesForces”. ''The Response of Worker Effort to Piece Rates: Evidence from the British Columbia TreePlanting Industry. pp. pp. 185199. Fixed Wages. pp.'' Journal of Human Resources. J. 34. 2000 Park. “LifeCycle Consumption and Hyperbolic Discount Functions”. 477490. 5992. Harry. 4. Management Science. 777793. 2010 Oyer. pp. Working Paper. 643667. pp. Stanford University. Management Science. Journal of Labor Economics. pp. and V. pp. 40. Quarterly Journal of Economics. Paul. Review of Economic Studies 35. and Robert Pollak.
Engle and D. Marketing Science. K. 1937 Steenburgh. 1987 Rust. Lorimer. 201206. 9991033. John. Special Issue on Enhancing Sales Force Productivity. 55.E. 2008 36 . in Handbook of Econometrics. ed.Rao. “Optimal Replacement of GMC Bus Engines: An Empirical Model of Harold Zurcher”. Chapter 14.. 115131. 28. 244264. 319342. Thomas. John. 1982 Zoltners. pp.. pp. pp. pp. “Some Empirical Evidence on Dynamic Inconsistency”. Paul. “Structural Analysis of Manufacturer Pricing in the Presence of Strategic Retailer”. Quantitative Marketing and Economics. 2001 Thaler. pp. 30823139. Prabhakant Sinha and Sally E. 8. 235256. Andris A. 1994 Samuelson. pp. 1990 Rust. Marketing Science. 2008 Sudhir. “A Note on Measurement of Utility”. Journal of Personal Selling and Sales Management. Richard H. Amsterdam: ElsevierNorth Holland. Summer. McFadden. pp. 6. Volume 4. “Sales Force Effectiveness: A Framework for Researchers and Practitioners”.. Ram C. Economic Letters. 9. “Compensating Heterogeneous Salesforces: Some Explicit Solutions”. by R.. “Effort or Timing: The Effect of LumpSum Bonuses”. Review of Economic Studies. 155161. Econometrica. pp. “Structural Estimation of Markov Decision Processes”. 4.
2 397. Sep Annual Bonus Base Commission Overachievement Commission * Dec Every month Dec These numbers are approximate for confidentiality reasons.8 42.1 1. Jun.3 % Achieving Quota ‐ Average Salary (USD) Average tenure (years) Average Q1 quota (000'USD) Average Q2 quota (000'USD) Average Q3 quota (000'USD) Average fullyear quota (000'USD) 51. Table 2: Descriptive statistics – Sales force under study Amount Size 348 $3.8 49.5%* paid in proportion to the revenue generated each month About 3%* paid in proportion to the total cumulative revenue surpassing the annual quota Mar.8 232.1 49.639.585 11.4 374.9 37 .Table 1: Firm's Compensation Plan Type Description Payment period Quarterly Bonus $1500 Awarded if quarterly revenue exceeds quarterly quota $4000 Awarded if annual revenue exceeds annual quota About 1.
188*** (0.565*** (0.01 *: p<0.554) 0.023* (0.024) 0.013) 0.022*** (0.005) 6.130) Yes *** Model 3 0.013) 0.004 (0.229) Yes Indirect Sales Sales person Fixed Effects Yes : p<0.302*** (0.15*** (2.007 (0.015) 0.021) 14.566*** (0.01 38 .063*** (2.014) 0.1 Table 4a: Parameter Estimates – Sales Response 0.002 (0.014) Last Month Sales Qtr 1st Month *Last Month Sales Qtr 1st Month*Last Month Sales* “Way off Target Last Qtr” Qtr 1st Month*Last Month Sales* “Not Way off Target” Monthly Allocated Quota 0.188*** (0.Table 3: Testing for Sales Substitution Across Months Model 1 0.029* (0.013) Model 2 0.021) 14.735 (5.003) Lagged annual quota Indirect sales Indirect sales*Lagged annual quota ***: p<0.
07) 0.52 (75.14 (17.41) 21.20) 87.33) 86.Table 4b: Parameter Estimates – Effort Policy Seg1 ρ0 ρ1(z1) ρ2(z1) ρ3(z1) ρ1(z2) ρ2(z2) ρ3(z2) ρ1(z1)*ρ1(z2) ρ1(z1)*ρ2(z2) ρ2(z1)*ρ1(z2) ρ1(z1)*ρ1(IS) ρ1(z2)*ρ1(IS) ρ1(z1)*ρ1(z2)*ρ1(IS) ρ1(z1)*ρ1(AQlag) ρ1(z2)*ρ1(AQlag) ρ1(z1)*ρ1(z2)*ρ1(AQlag) ρ1(T) ρ2(T) ρ3(T) 18.12 (11.37) 6.000) 39 Seg2 74.52 (17.68) 9.61) 113.10*** (44.07 (14.02 (0.11 (2.03) 0.04) 18.53*** (6.95) 0.34*** (4.62) 20.002*** (0.67) 8.90) 0.85) 42.98 (5.04** (0.32) 23.05.38) 0.81*** (36.83*** (40.01) 0.59*** (117.00*** (2.79) 110.14) 241.18) 43.54) 31.05) 0.27*** (0.00 (181.42 (29.54 (17.05*** (0.31) 16.40) 10.001) Seg3 263.003) ***: p<0.05 (0.08) 8. **: p<0.01.85) 148.87) 0.004 (0.03) 9.28 (55.77) 9.28) 312.13 (18.90 (10.06*** (0.26*** (38.00) 22.23) 15.36** (137.60 (19.10*** (8.11** (44.76*** (3.1 .74) 146.91*** (21.02) 0.50) 15.27 (18.89) 11.001 (0.04) 0.02) 0.01 (0.09) 141.73*** (44.41 (3.91*** (18.88) 0.02 (0.95) 1.53) 89.62) 0.39) 114.03** (0.87 (13.01) 0.53 (1.26) 0.52** (0.09* (4.93) 62.00** (8.63) 20.98 (82.09) 6.09) 458.67 (22.00*** (89.07*** (0.28) 16.42*** (55. *: p<0.94) 102.01) 0.14) 0.39*** (63.04 (0.23** (32.63 (26.
30 1.4 0.35 Table 5: Descriptive Characteristics of Segments segment1 Share Tenure* Achieve quarterly quota .1167 0.54 0.1200 0.1100 0.1243 0.1181 0.1300 0.1207 0.1140 0.57 1.1127 0.1382 0.1074 0.Q2 Achieve quarterly quota .72 seg3 279.1192 0.1121 0.1347 0.1242 0.1117 0.97 0.1336 0.1107 0.5 0.1181 0.49 0.6 0.0 segment3 0.1193 0.1168 0.1185 0.1167 0.1391 0.1328 β 0.1300 0.91 0.4 130.21 11.Q3 Achieve annual quota Average annual quota** Average December revenue** 0.Q1 Achieve quarterly quota .1251 0.5 0.7 273.32 11.615.1475 0.1148 0.9 0.1287 0.46 0.95 0.1220 0.62 0.1134 0.1108 0.1402 0.1167 0.1172 0.1223 0.1428 0.1149 0.1258 0.55 0.1202 0.2 0.1145 0.1149 0.1098 0.1098 0.1180 0.201.93 0.1101 0.92 0.1084 0.1124 0.1318 0.1378 0.61 seg2 143.7 0.1132 0.1203 0.8 0.94 0.47 12.9 1 40 .Table 4c: Revenue Shock Distribution – Standard Deviation seg1 Sigma 81.1367 0.1450 0.1266 0.2 segment2 0.1198 0.363.5 0.38 0.1219 0.1203 0.1255 0.53 0.64 2.1103 0.31 0.1249 0.58 0.1270 0.1182 0.99 0.1 *Tenure is measured in years **Average quotas and revenues are indicated in USD(K) Table 6: Optimal Discount Factor – Model Fit Mean Absolute Percentage Error by Discount Factors δ 0.1175 0.96 0.1103 0.98 0.1150 0.1109 0.7 559.1104 0.1241 0.
Only Pure Commissions (adjusted to equal payout with bonus) 2a.092 (0.8% 3.0020) Segment 2 Disutility 0.038) Risk Aversion 0.01 Without Forward Looking Behavior 0.0002) ***p<0. Only Pure Commissions 1b.0000 (0. No Bonus (Only Commissions + Overachievement Commission) 2b.0003) 0.199 (0.3% 4.0001 (0.0% 4.119 (0.002) 0.005) Risk Aversion 0.0005) Segment 3 Disutility 0.0005) 0.6% 41 5a.0004) Table 8: Impact of Alternative Bonus Plans on Sales Revenues Counterfactual 1a. No overachievement commissions 4a.006) 0. No Bonus (Commissions adjusted to equal payout with bonus) 3.8% 9.0001 (0.0001 (0. Annual Bonus split into Quarterly and Annual Bonus Change in `Revenues 20. Remove quarterly bonus .Table 7: Parameter Estimates With Forward Looking Behavior Segment 1 Disutility 0.0001 (0.240 (0. Cumulative Annual Quota replaced with quarterly quota 4b.061 (0.2% 1.3% 1.5% 13.005) 0.048 (0.002) Risk Aversion 0.0000 (0.
0% Figure 1: Plots of Incentive Compensation Schemes Plan A: Pure Commission Plan B: Pure Bonus Plan C: Commission at Quota Earnings Earnings Sales Sales Earnings Sales Plan D: Commission with Floor and Ceiling Plan E: Commission+Bonus Plan F: Commission+Bonus +Overachievement Commission Earnings Earnings Sales Sales Earnings Sales 42 .1% 2.0% 12.Table 9: Impact of Alternative Bonus Plans on Sales Revenues by Segment % decrease from different components Pure commission Without overachievement Without quarterly bonus Seg1 17.6% 4.9% 7.0% 10.5% Seg3 21.0% Seg2 21.4% 17.
e*=5 50 0 50 0 2 4 6 8 10 50 0 50 0 2 4 6 8 10 d=1. r=10.5 50 0 50 0 100 150 2 4 6 8 10 0 0 50 100 2 4 6 8 10 0 50 100 0 2 4 6 8 10 50 d=2. B=30 S=0.5 3 b) Sales ‐ Indirect Sales Force Normalized Sales Feb Sep 2.5 2 1.B=30 e*=10 200 0 0 2 4 6 8 10 d=1. r=10. r=20.Figure 2a: How Quotas and Bonus Serve as stretch goals d=1.5 1 0.5 0 USD (K) 150 100 50 0 Feb Aug Dec Aug Sep Jul Jan Jan Jul May May Nov Mar c) Percentage of Fiscal year‐ends 80 70 Percentage 60 50 40 30 20 10 0 Feb May Aug Sep Jan Jun Jul Mar Mar 43 Nov Dec Apr Oct Nov Dec Jun Oct Jun Apr Apr Oct .B=0 e*=10 Figure 2b: Effort as a Function of Distance to Quotas d=2. r=10. e*=2. r=10. e*=5 100 50 d=2. B=30 S=7.B=0. B=30 S=5.r=10. e*=3 Figure 3: Sales / IndirectSales / Fiscal YearEnds a) Average Sales 300 250 200 3.
29 44 . pwidth = .18.2. degree = 0.6 .6 . bandwidth = .23.2 % Quarterly Quota Sold by Feb 1. bandwidth = .2 .17. degree = 0.4 .8 1 1.8 1 % Quarterly Quota Sold by Apr 1.26 Figure 4b: Sales and Percentage Quota Achieved – PreBonus Months Feb Sales/Feb Quota 1 2 3 0 0 .8 1 1.5 0 0 .8 1 % Quarterly Quota Sold by Jan 1.31 kernel = epanechnikov.2 kernel = epanechnikov. bandwidth = . pwidth = .4 .6 .25 0 0 .2 kernel = epanechnikov. pwidth = .8 1 1.2 0 0 Nov Sales/Nov Quota 1 2 3 Aug Sales/Aug Quota 1 2 3 .23 kernel = epanechnikov. degree = 0.4 .4 kernel = epanechnikov.4 .4 .6 .2 % Quarterly Quota Sold by May 1.4 .6 . degree = 0. pwidth = . bandwidth = .16.2 .4 kernel = epanechnikov.2 % Quarterly Quota Sold by Aug 1. pwidth = .Figure 4a: Sales and Percentage Quota Achieved – Bonus Months Mar Sales/Mar Quota 1 2 3 0 0 .2 0 0 May Sales/May Quota 1 2 3 .17.33. degree = 0.8 1 % Annual Quota Sold by Oct 1.2 .4 0 0 Dec Sales/Dec Quota 1 2 3 Sep Sales/Sep Quota 1 2 3 .2 .35 kernel = epanechnikov.4 0 0 Jun Sales/Jun Quota 1 2 3 .2 .6 . pwidth = . degree = 0.6 .27 kernel = epanechnikov.4 .2 % Annual Quota Sold by Nov 1. pwidth = .2 . bandwidth = . bandwidth = .4 .2 .2 . bandwidth = . degree = 0. bandwidth = .6 .8 1 1. degree = 0. pwidth = .21.8 1 % Quarterly Quota Sold by July 1.
65 0.30 % QQ % AQ Figure 5b: The Effect of Tenure on Effort 120 100 80 The Effect of Tenure Effort 60 40 20 (20) 1 2 3 4 5 6 7 8 9 10 11 12 13 Tenure (years) Figure 6: Simulated Revenue & Effort– Static vs.60 0.60 0.30 0.20 1.35 0.85 0. Dynamic 45 .15 0.10 1.80 400 350 300 December Effort Effort 250 200 150 100 50 0.20 0.95 1.50 0.Figure 5a: Effort Policy by Segment as a Function of % Quota March 250 200 150 100 50 0.05 1.25 1.70 0.50 0.00 1.45 0.25 0.75 0.05 0.15 1.65 0.70 0.75 0.00 0.10 0.55 0.90 0.55 0.40 0.80 0.
Revenue 250 200 USD (K) USD (K) 150 100 50 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec 140 120 100 80 60 40 20  b.a. Effort Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Dynamic Myopic Dynamic Myopic Figure 7: Effect of Overachievement Commission 150 130 USD (K) 110 90 70 50 30 Nov Jan Mar Apr Jun Aug Feb Sep Oct May Dec Jul Met quota Didn't meet quota Figure 8: Effect of Quarterly Quotas 120 110 100 USD (K) 90 80 70 60 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Current plan Without quarterly bonus 46 .