MKSC 2013 0815

This article was downloaded by: [134.7.34.
232] On: 29 October 2018, At: 08:43

Publisher: Institute for Operations Research and the Management Sciences (INFORMS)
INFORMS is located in Maryland, USA
Marketing Science
Publication details, including instructions for authors and subscription information:
http://pubsonline.informs.org
Do Bonuses Enhance Sales Productivity? A Dynamic

Structural Analysis of Bonus-Based Compensation Plans
Doug J. Chung, Thomas Steenburgh, K. Sudhir
To cite this article:

Doug J. Chung, Thomas Steenburgh, K. Sudhir (2014) Do Bonuses Enhance Sales Productivity? A Dynamic Structural Analysis of
Bonus-Based Compensation Plans. Marketing Science 33(2):165-187. https://doi.org/10.1287/mksc.2013.0815
Full terms and conditions of use: http://pubsonline.informs.org/page/terms-and-conditions
This article may be used only for the purposes of research, teaching, and/or private study. Commercial use
or systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisher
approval, unless otherwise noted. For more information, contact permissions@informs.org.
The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitness
for a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, or
inclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, or
support of claims made of that product, publication, or service.
Copyright © 2014, INFORMS
Please scroll down for article—it is on subsequent pages
INFORMS is the largest professional society in the world for professionals in the fields of operations research, management
science, and analytics.
For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org
Vol. 33, No. 2, March–April 2014, pp. 165–187
ISSN 0732-2399 (print) ISSN 1526-548X (online) http://dx.doi.org/10.1287/mksc.2013.0815
© 2014 INFORMS
Do Bonuses Enhance Sales Productivity?

A Dynamic Structural Analysis of Bonus-Based
Compensation Plans
Doug J. Chung
Harvard Business School, Boston, Massachusetts 02163, dchung@hbs.edu
Thomas Steenburgh
Darden Graduate School of Business, University of Virginia, Charlottesville, Virginia 22906,
steenburght@darden.virginia.edu
K. Sudhir
Yale School of Management, New Haven, Connecticut 06520, k.sudhir@yale.edu
W e estimate a dynamic structural model of sales force response to a bonus-based compensation plan. This
paper provides substantive insight into how different elements of the compensation plan enhance pro-
ductivity. 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) quar-
terly bonuses help improve performance of the weak performers by serving as pacers to keep the sales force
on track in achieving its annual sales quotas. The paper also introduces two main methodological innovations
to the marketing literature: First, we implement empirically the method proposed by Arcidiacono and Miller
[Arcidiacono P, Miller RA (2011) Conditional choice probability estimation of dynamic discrete choice models
with unobserved heterogeneity. Econometrica 79(6):1823–1867] to accommodate unobserved latent-class hetero-
geneity using a computationally light two-step estimator. Second, we illustrate how discount factors can be
estimated in a dynamic structural model using field data through a combination of (1) an exclusion restric-
tion separating current and future payoff and (2) a finite-horizon model in which there is no forward-looking
behavior in the last period.
Keywords: sales force compensation; bonuses; quotas; dynamic structural models; two-step estimation;
discount factors
History: Received: July 24, 2011; accepted: August 4, 2013; Preyas Desai served as the editor-in-chief and
Miguel Villas-Boas served as associate editor for this article. Published online in Articles in Advance
November 7, 2013.
1. Introduction Relative sales force expenditure elasticity is useful

Personal selling is one of the most important ele- in determining the relative effectiveness of different
ments of the marketing mix, especially in the con- instruments in the marketing mix, but little insight is
text of business-to-business (B2B) firms. An estimated given on how to design a sales force compensation
20 million people work as salespeople in the United plan, which is widely understood to be the primary
States (Zoltners et al. 2008). Sales force costs average tool by which firms can induce their sales forces to
about 10% of sales revenues and as much as 40% of exert the optimal levels of effort and thus to optimize
sales revenues for certain industries (Heide 1999). In the use of sales force expenditures.
the aggregate, U.S. firms spent more than $800 billion A compensation plan can consist of many compo-
on sales forces in 2006, a sum three times larger than nents: salary, commissions, and bonuses on achiev-
advertising spending (Zoltners et al. 2008). ing a certain threshold of performance called quotas.
Marketing researchers routinely create response Figure 1 shows a variety of compensation plans that
models for marketing mix instruments such as price, include combinations of these components. Which of
sales promotion, and advertising. Meta-analyses of these different types of contracts should a particu-
various research studies estimate that the sales force lar firm offer to its sales force to maximize prof-
expenditure elasticity is about 0.34 (Albers et al. 2010), its? What combination of salary, commission, and/or
relative to about 0.22 for advertising (Assmus et al. quota-based bonuses should one use? Having chosen
1984) and about −2.62 for price (Bijmolt et al. 2005). the compensation components, what should be the
165
Chung, Steenburgh, and Sudhir: Do Bonuses Enhance Sales Productivity?
166 Marketing Science 33(2), pp. 165–187, © 2014 INFORMS
Figure 1 Types of Incentive Compensation Schemes
Plan A: Pure commission Plan B: Pure bonus Plan C: Commission at quota

Earnings
Earnings
Earnings
Sales Sales Sales
Plan D: Commission with Plan E: Commission + Bonus Plan F: Commission + Bonus

floor and ceiling + Overachievement commission
Earnings
Earnings
Earnings
Sales Sales Sales
specific parameters for commission rate, quotas, and of quotas and bonuses relative to straight linear com-
bonus levels? Furthermore, what is the right frequency mission plans.
for quota targets? For example, should there be quar- We begin with a discussion of the relevant theoreti-
terly or annual quotas? A firm needs to understand cal literature. Using the principal-agent framework of
how the sales force will respond to different elements Holmstrom (1979), Basu et al. (1985) and Rao (1990)
to develop an appropriate compensation plan. find that the combination of salary and commission
Thus, this paper has two substantive goals: First, it (usually nonlinear with respect to sales) is optimal.
seeks to gain insight into how a firm should design In this light, quota-bonus plans can be seen as an
its compensation plan. For example, should a firm approximation to a continuous nonlinear plan that
offer quotas and bonuses in addition to commis- also takes into account heterogeneity in territory sales
sions? Second, we aim to determine how often quotas potential. However, under the assumption of linear
should be set and bonuses paid. For example, should exponential utility and normal errors, Holmstrom and
a firm implement a monthly, quarterly, or annual Milgrom (1987) and Lal and Srinivasan (1993) show
bonus? Should it offer a quarterly bonus in addi- that a linear commission scheme can achieve the first-
tion to an annual bonus? In the education literature, best outcome. Yet why do we see quota-bonus plans?
researchers have argued that frequent testing leads to
Raju and Srinivasan (1996) suggest that even though a
better performance outcomes (Bangert-Drowns et al.
commission-over-quota plan may not be theoretically
1991). Can quarterly quotas serve a similar role
optimal, it provides the best compromise between
to improve outcomes? As in education, where fre-
efficiency and ease of implementation. Others argue
quent exams help students to be better prepared for
that quota-based plans offer high-powered incentives
the comprehensive final exam, frequent quota-bonus
plans may serve as a mechanism to keep the sales that can motivate salespeople to work harder (e.g.,
force motivated to perform well in the short run so Darmon 1997). Park (1995) and Kim (1997) demon-
as to be within striking distance of the overall annual strate that a quota-bonus plan may lead to the first-
performance quota. best outcome, but in their framework, a quota-bonus
Quotas and bonuses are widely used by firms. plan is just one of many possible plans that can lead to
According to Joseph and Kalwani (1998), only about first-best outcomes. Oyer (2000) shows that when par-
24% of firms use a pure commission-based plan; the ticipation constraints are not binding, a quota-bonus
rest used some form of quotas. As per the 2008 Incen- plan with linear commissions beyond quotas can be
tive Practices Research Study by ZS Associates, 73%, uniquely optimal because it can concentrate the com-
85%, and 89% of firms in the pharma/biotech, med- pensation in the region of effort where the marginal
ical devices, and high-tech industries, respectively, revenue from effort minus the cost of compensation
use quota-based compensation (Training 2008). Yet, is maximized.
despite the ubiquity of quota-based compensation, In terms of empirical work, Ferrall and Shearer
there is considerable controversy in both the theoret- (1999) and Paarsch and Shearer (2000) estimate static
ical and empirical literatures about the effectiveness structural models of worker behavior given linear
Marketing Science 33(2), pp. 165–187, © 2014 INFORMS 167
contracts. Oyer (1998) is the first to empirically inves- estimation approaches has recently gained popular-
tigate quota-based plans. Using aggregate sales across ity (Hotz and Miller 1993, Bajari et al. 2007) because
different industries in different quarters, he con- of their ease of computation relative to traditional
cludes that quota-based plans encourage salespeople nested fixed-point estimation approaches (e.g., Rust
to maneuver the timing of orders; this negative effect 1987), their use in empirical applications has been lim-
overwhelms the positive benefit of quotas. Using ited by their inability to accommodate unobserved
reduced-form analysis of individual-level data, Steen- heterogeneity. Arcidiacono and Miller (2011) propose
burgh (2008) finds that the net improvement in rev- an algorithm that accommodates latent-class hetero-
enues from effort dominates the inefficiencies induced geneity within the two-step framework. Our paper
by intertemporal dynamic considerations and shows introduces this idea to the marketing literature and
that an aggregate analysis might have led to the oppo- illustrates its implementation with an empirical appli-
site conclusion similar to that of Oyer (1998). cation that accommodates unobserved heterogeneity
Copeland and Monnet (2009) estimate the first using two-step dynamic structural estimation.2
dynamic structural model of worker productivity in a Second, unlike Misra and Nair (2011)—and of
check-sorting environment with nonlinear incentives. broader importance to the dynamic structural mod-
Unlike sales force productivity, where we observe eling literature—we estimate rather than assume dis-
only aggregate sales, the outcomes associated with count factors. It is well known in the literature
every processed check are observed in their envi- on dynamic structural models that discount fac-
ronment; hence, they have a qualitatively different tors cannot be identified in standard applications
model. Our paper shares many similarities with a because there are no instruments that provide exclu-
recent paper by Misra and Nair (2011), who also sion restrictions across current and future period pay-
estimate a dynamic structural model of sales force offs (Rust 1994). Hence the standard approach is to
compensation, although they use a very different assume discount factors. We illustrate how the bonus
quota-compensation scheme. In contrast to our focus setting provides us exclusion restrictions allowing us
on quotas with bonuses (plan F in Figure 1), Misra to estimate discount factors—thus demonstrating how
and Nair analyze quotas with floors and ceilings on field data can indeed be used to estimate discount
commissions (plan D in Figure 1). They conclude that factors.3 Since bonus payoffs occur only at the end
quotas reduce performance. This is because of two of each quarter or year, in the nonbonus periods, the
characteristics of their quotas: First, the quota ceil- probability of achieving the quota and receiving a
ing (beyond which salespeople receive zero additional bonus will not affect the current payoff, only future
compensation) limits the effort of the most produc- payoffs. Only a forward-looking person (i.e., one with
tive salespeople, who would normally have exceeded a nonzero discount factor) would be concerned with
that ceiling. Second, the company followed an explicit whether she is close to the quota in nonbonus peri-
policy of ratcheting quotas based on past productiv- ods. We demonstrate through reduced-form evidence
ity. This reduces salespeoples’ incentives to work hard that such behavior exists in the data, and then we
in any given period, because hard work is penalized exploit this exclusion restriction to identify the dis-
through higher future quotas. In contrast, we find that count factor. In addition, the finite-horizon nature of
quotas coupled with bonuses enhance performance. the sales force problem implies that, in the last period,
In the plan we consider, the company offers extra the model reduces to a static model, for which it is
overachievement commissions for exceeding quotas well known that utility can be identified. This fur-
and uses a group quota updating procedure that min- ther facilitates identification of the discount factor. Yao
imizes ratcheting effects. Thus, these two papers offer et al. (2012) apply our idea to a cellphone usage con-
complementary perspectives that enhance our under- text to identify discount factors. In their setting, how
standing of how quotas impact performance.
From a methodological perspective, our paper intro- number of observations per individual. More importantly, the
approach requires that salespeople exert effort equally across all
duces two key ideas to the marketing literature. First,
customers—an assumption they show is valid in their data but
we accommodate latent-class heterogeneity within the unlikely to hold in general. Our latent-class approach works in the
two-step conditional choice probability (CCP) estima- more common situation where there are limited observations per
tion framework—an issue that has been an economet- individual.
ric challenge for the literature for nearly two decades. 2
Finger (2012) and Beauchamp (2012) are two concurrent work-
Misra and Nair (2011) avoid the unobserved hetero- ing papers implementing Arcidiacono and Miller’s approach in
geneity issue by estimating each salesperson’s util- economics.
3
ity function separately.1 Although the use of two-step Use of exclusion restrictions to empirically infer discount factors
is important because, prior to this paper, the conventional wisdom
was that field data cannot be used for discount factor identification
1
This is similar to estimating individual-level utility functions in and that one needs to use surveys (e.g., conjoint data as in Dubé
conjoint analysis or scanner panel data, when there are a large et al. 2012).
close the user is to exceeding his or her monthly quota manager should have greater faith in the counterfac-
serves as an exclusion restriction, much like in our tuals based on parameters that were estimated from
setting. Similar to the last period of our data, where observed responses to different elements of the com-
the model reduces to a static model, they have data pensation plan.
for a period in which pricing is linear and therefore The rest of the paper is organized as follows. Sec-
reduces to a static model. tion 2 discusses the institutional details of the com-
There are three specific modeling and estimation pensation plan at the firm, provides a numerical
challenges in the structural estimation of response to example that offers insights into how bonuses induce
compensation plans, especially those with quotas and effort, and provides some model-free evidence that
bonuses. First, we do not observe the effort of the facilitates model building. We present the model and
sales force, only the outcome of the agent’s effort (i.e., the estimation methodology in §§3 and 4, respec-
sales), which is correlated with effort. This requires tively. Section 5 discusses the estimation results and
a modeling assumption on the link between sales the counterfactual analysis. Section 6 concludes.
and effort. The identification of effort poses particular
challenges in our application as a result of seasonal- 2. Institutional Details and
ity and potential sales substitution across quarters.4 Descriptive Analysis
We discuss these issues in §2.3. Second, compensa- We first describe the details of the bonus-based com-
tion plans do not change over time. Here, we draw pensation plan, followed by a numerical example to
on an empirical insight from Steenburgh (2008) for clarify how bonuses can help motivate the salesper-
identification. In any given period, a sales agent’s son and induce intertemporal effort that assists us
optimal effort depends on her state—how close the with identification. We then provide model-free evi-
person is to achieving her quota. A sales agent may dence of forward-looking behavior, seasonality, and
find it optimal to reduce her effort when she is close the absence of sales substitution across quarters.
or very far from achieving the quota, but she may
2.1. The Compensation Plan
stretch herself to reach the quota when she has a mod-
The focal firm under study is a highly regarded multi-
erate chance of achieving it. Thus changes in sales
national Fortune 500 company that sells durable office
in response to changes in the agent’s state (distance
products primarily using its own direct sales force.
to quota) within and across agents facilitate identifi-
Each sales agent is given an “exclusive” territory; the
cation. Finally, quotas and bonuses induce dynamic
firm traditionally does not encourage group work or
forward-looking behavior; an agent choosing to exert
team cooperation among the sales force. The firm also
effort has to be concerned not just with the current has an indirect sales force through “rep” firms who do
payoff but with the effect of that effort on the likeli- not compete with the direct sales force. They are paid
hood of obtaining a bonus in the future. This requires purely on commission, unlike the regular sales force.5
a dynamic structural model. Our analysis focuses on sales performance data
We estimate a dynamic structural model of sales from 348 salespeople from the regular sales force dur-
force response to various features of the compensation ing the three-year period 1999–2001. The firm’s com-
plan using sales force output and compensation data pensation structure follows the pattern in plan F of
from a Fortune 500 firm that sells office durable goods. Figure 1, and the details of the compensation schedule
This firm uses plan F in Figure 1. In addition, bonuses for the period of analysis are described in Table 1. We
are provided at two different frequencies: quarterly provide descriptive statistics of the data in Table 2.
and annual. As the compensation structure of the Every month, salespeople receive a fixed monthly
focal firm features almost all dimensions in typically salary (on average, $3,585) and a commission of 1.5%
used compensation plans, we observe how the sales of the revenues generated in that month. In the first
force responds to the plan’s different dimensions. This three quarters, a quarterly lump-sum bonus of $1,500
“rich” plan provides us with two key benefits: First, is paid if the quarterly quotas are met. At the end of
the presence of bonuses helps us identify and esti- the year (i.e., end of the fourth quarter), an annual
mate discount factors. Second, even though in theory lump-sum bonus of $4,000 is paid if the annual quota
one can perform counterfactuals of any type of com- is met. Furthermore, an overachievement commis-
pensation plan if able to estimate structural param- sion of 3% is paid for any excess revenues beyond
eters (other than discount factors) for a salesperson the annual quota. There are no caps on revenues for
with a less rich compensation plan, an analyst or which an agent could obtain commissions or over-
achievement commissions. Overall, for a salesperson
4
that meets all quotas, the salary component will be
This issue has parallels in empirical channel response models.
For example, Sudhir (2001) makes an inference about manufacturer
roughly 50% of total compensation.
actions (wholesale prices) from the observed retail price and sales
5
to infer competition between manufacturers. Such rep firms are the focus of Jiang and Palmatier (2009).
Table 1 Firm’s Compensation Plan
Type Description Payment period
Quarterly bonus $1,500 awarded if the quarterly revenue exceeds March, June, September
quarterly quota
Annual bonus $4,000 awarded if the annual revenue exceeds December
annual quota
Base commission About 1.5%a paid in proportion to the revenue Every month
generated each month
Overachievement About 3%a paid in proportion to the total cumulative December
commission revenue surpassing the annual quota
a
These numbers are approximate for confidentiality reasons.
In building annual and quarterly quotas for its Table 3 Testing for Ratcheting Effects
sales force, the company uses internal metrics called
Model 1 Model 2 Model 3
“monthly allocated quotas” (based on expected
monthly revenues, given seasonality, and territorial Dependent variable 4AQ y − AQ y −1 5/AQ y −1 log4AQ y 5 log4AQ y 5
characteristics), though these are not used for com- I4Sales y −1 ≥ AQ y −1 5 0.0364 0.0033
pensation. We do not use these quotas for our mod- (0.0229) (0.0299)
eling and estimation, but we use them to benchmark log4%AQ y −1 5 0.0347
performance in the reduced-form analysis. (0.0196)
The most important element in performance eval- Year fixed effects Yes Yes Yes
uation within the firm is the annual quota; i.e., the Salesperson fixed effects NA Yes Yes
firm views a salesperson as having a successful year
Notes. Standard errors are shown in parentheses. y represents each year in
if the annual quota is met. From Table 2, we see that this analysis.
salespeople meet their annual quota roughly 50% of
the time. All quotas, including the quarterly quotas,
regress the logarithm of the annual quota on whether
are updated annually and reset in January of the fol-
a salesperson met her quota in the previous year, this
lowing year. However, managers at the firm informed
time controlling for salesperson fixed effects. Again,
us that they were sensitive to the fact that ratcheting
we find no significant effect. Finally, in Model 3, we
quotas based on individual performance could lead
regress the logarithm of quota on the logarithm of
the sales force to purposefully reduce performance.
the previous year’s performance relative to the annual
To avoid such an adverse impact on sales, the quota
quota. Again, we find no significant effects. Thus,
adjustment year after year was done based only on
consistent with the intuition of the firm’s managers,
group performance, where each individual’s current
who sought to avoid the negative impact of ratch-
performance would have minimal direct impact on
eting on employee performance, we find no direct
her future quota. Nevertheless, we test for statistical
statistical evidence of ratcheting. For this application,
evidence of ratcheting in the data. Table 3 reports the
results of the tests. In Model 1, we regress the percent- we therefore abstract away from quota ratcheting and
age increase in the annual quota on whether the sales- treat each salesperson’s quarterly and annual quotas
person met her quota in the previous year. We find as exogenous. To the extent that a salesperson’s quota
no significant effect of meeting the previous year’s is partly updated based on her own performance,
quota on the current year’s quota. In Model 2, we there are still potential ratcheting effects that may be
detectible with longer panel data.6
Table 2 Descriptive Statistics 2.2. Numerical Example

No. of salespeople 348
Before building an empirical model of salesperson’s
Average salary (US$) 3,585 response to a bonus-based compensation scheme, we
Average tenure (years) 11.8 provide a numerical example that illustrates how a
quota-bonus scheme can be more effective than pure
Average quota % achieving Average sales
(US$ ’000) quota (US$ ’000) commissions in generating more effort. The exam-
ple will also show that a person’s distance to quota
Quarter 1 23204 51.1 27300
Quarter 2 37402 49.8 44500
6
Quarter 3 39701 42.8 40700 The assumption of no ratcheting allows us to make the finite-
Quarter 4 — — 56506 horizon argument for identification of the discount factor discussed
in §4.3. The exclusion restriction argument for identification will be
Annual 1163903 49.9 11 69006
valid even if we allow for ratcheting.
can induce variations in optimal effort (and gener- can be a way to approximate the optimal nonlinear
ated sales)—providing intuition for our identification incentive plan.
strategy. Second, we show how distance to quota induces
Let the utility function of the salesperson that variation in effort. Let d = 2, r = 10, Q = 10, and
trades off effort (e) and income from sales (s) and who B = 30. Consider three scenarios of distance to quotas:
has sold S units at the beginning of the new period be S = 0 (far away from quota), S = 5 (moderately close
to quota), and S = 7 (close to quota). Figure 2(b)
U 4s1 e1 S5 = −de2 + rs + BI8s+S≥Q9 1 shows that the optimal effort levels are e∗ = 205, 5,
and 3, respectively, for the three cases; the salesper-
where −d is the disutility parameter and r is the com- son exerts maximum effort at S = 5, when moderately
mission rate (d > 01 r > 0), and B is the bonus for far away from quota, all else being equal. We use this
reaching quota (Q). For illustration, let us assume a variation of optimal levels of effort (thus, sales) within
direct match between sales and effort; i.e., s = e. agents across time to identify their preferences in our
First, we illustrate the effectiveness of bonuses main model.
using a static model. For simplicity, assume S = 0.
In the pure commission case with no bonus, where 2.3. Model-Free Analysis
d = 1, r = 10, and B = 0, the optimal effort is e∗ = 5. We consider three features of the data that inform
In the bonus case, where Q = 10 and B = 30, the opti- model development. First, we provide evidence of
mal effort is higher, at e∗ = 10, and the compensation forward-looking behavior induced by bonuses and
cost to the firm is $130. To achieve the same level hence the need to develop a dynamic model. Sec-
of sales and effort (e∗ = 10) from a pure commission ond, we demonstrate the presence of seasonality in
plan, the commission rate r has to increase to 20 and the data, which therefore requires us to accommodate
costs more for the firm, at $200. Figure 2(a) illustrates this in the model. Third, we show that sales substi-
these results graphically. Thus, the quota-bonus plan tution across quarters by sales agents appears to be
in this context induces more effort and sales for the limited in our data; we therefore abstract away from
same level of compensation. We are not claiming that modeling the timing of sales bookings by sales agents.
a bonus-based plan is the optimal compensation plan. 2.3.1. Forward-Looking Behavior. As discussed
Through the example above, we are simply illustrat- previously, bonuses provide an exclusion restriction in
ing that a nonlinear plan can be more efficient than a that they do not impact current payoffs, only future
linear plan, suggesting that the optimal compensation payoffs. To the extent that a sales agent’s perfor-
plan is possibly nonlinear and that the bonus plan mance is affected by variables relating to proximity
Figure 2(a) How Quotas and Bonus Serve as Stretch Goals
d = 1, r = 10, B = 0, e* = 5 d = 1, r = 10, B = 30, e* = 10 d = 1, r = 20, B = 0, e* = 10

40 40 120
100
80
Utility
20 20
Utility
Utility
60
40
0 0 20
0 2 4 6 8 10 0 2 4 6 8 10 0
–20 –20 0 2 4 6 8 10
Effort Effort Effort
Figure 2(b) Effort as a Function of Distance to Quotas
d = 2, r = 10, B = 30, d = 2, r = 10, B = 30 d = 2, r = 10, B = 30

S = 0, e* = 2.5 S = 5, e* = 5 S = 7, e* = 3
50 100
50
0 50
0 2 4 6 8 10 0
Utility
Utility
Utility
0 2 4 6 8 10
–50 0
0 2 4 6 8 10
–50
–100 –50
–150 –100 –100

Effort Effort
Effort
Figure 3 Fraction of People Achieving Quota in December Table 4 Sales Performance in November
1.2 Estimate
%AQ < 005 %AQ > 005

1.0
Intercept 0005∗∗∗ 0006∗∗∗
Probability of achieving annual quota
4000185 40001235
0.8 %AQ 0006 0004∗∗∗
4000495 4−0001265
0.6 Note. Standard deviations are shown in

parentheses.
∗∗∗
p < 0001.
0.4
Two key elements stand out from Figure 4(a). First,

0.2 across the board there is little reduction in effort when
salespeople are close to achieving quota, most likely
as a result of the overachievement commission rate.
0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3
Second, there is a steady increase over time in the
%AQ at the end of October %QQ and %AQ threshold beyond which salespeople
–0.2 reach their monthly targets. The threshold is about
25% in March, 35% in June, 45% in September, and
to bonuses, this is evidence of forward-looking close to 70% in December. Early in the year, even if
behavior.7 But proximity to bonus quota will impact below targets, salespeople still have hopes of receiv-
performance only if agents have a reasonable chance ing a large annual bonus by working hard. As one
of making the quota. Figure 3 shows the graph of the gets closer to year-end, the chances of reaching the
probability of reaching the annual quota, conditional quota become less likely, and salespeople respond by
on the cumulative fraction of annual quota (%AQ) reducing effort even at higher levels of %AQ and
achieved by November. It is clear that there is very %QQ. Since annual bonuses should have no impact
little chance of achieving quota if %AQ < 005. We on current payoffs in March, June, or September,
therefore divide agents by their state %AQ < 005 and only on future payoffs, this is suggestive of forward-
%AQ > 005 to test whether the state affects sales and looking behavior.
estimate regressions on sales performance in Novem- The next set of graphs presented in Figure 4(b)
ber as a function of their states. Table 4 reports the shows the same relationship in the pre-bonus months
results of the regressions. Consistent with forward- (February, May, August, and November) and provides
looking behavior, the state %AQ is significant only for additional evidence for forward-looking behavior. In
agents with %AQ > 005.8 February, May, and even August, at all levels of %QQ,
For additional evidence of forward-looking behav- the salesperson on average sells above the monthly
ior, we show scatterplots and the best-fitting non- allocated quota. This is because hard work (and some
parametric smoothed polynomial (and its 95% con- good luck in the form of positive sales shocks) may
fidence interval) of sales revenues normalized by give the salesperson a reasonable chance of attaining
monthly allocated quotas in the quarterly bonus the smaller quarterly targets. However, in November,
months (March, June, September, and December) only at a very high level of %AQ does the salesperson
against the percentage of quota attained by the pre- sell above the monthly allocated quota, because one
vious month in Figure 4(a). For March, June, and has a very limited chance of closing the large gap in
September, the x axis is the percentage of quarterly just two months. In the pre-quarterly bonus months,
quota completed (%QQ), whereas for December, the the immediate future quarterly bonus impacts behav-
x axis is the percentage of annual quota completed ior, even though it has no impact on the current pay-
(%AQ). The vertical line shows the %QQ and %AQ at off; this indicates more conclusive forward-looking
which salespeople, on average, achieve their monthly behavior.
allocated quotas. This evidence leads to a natural question: Should
the large annual bonus be split into a quarterly bonus
7
(as in other months) and an annual bonus? The quar-
A similar argument is made in providing evidence of forward-
terly bonus can prevent salespeople from giving up in
looking behavior by students in the textbook market by Chevalier
and Goolsbee (2009). November, even if they do not have a chance of reach-
8
The qualitative conclusions are robust in the range of thresholds ing the annual quota. But with such a quarterly quota,
of %AQ from 0.4 to 0.6. Note that if there were no overachievement early in the year, agents may have limited incentive to
commission, agents very close to the quota may reduce their effort. work hard after reaching quarterly quotas. How these
Figure 4(a) Sales and Percentage Quota Achieved—Bonus Months
3 3
Mar sales/Mar quota
Jun sales/Jun quota

2 2
1 1
0
0
0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
% quarterly quota sold by Feb % quarterly quota sold by May
3 3
Sep sales/Sep quota
Dec sales/Dec quota

2 2
1 1
0 0
0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
% quarterly quota sold by Aug % annual quota sold by Nov
two issues trade off is an empirical question that we force. There are clear peaks at the end of each quar-
subsequently address in the counterfactual analysis. ter. These peaks could be due to either seasonality
2.3.2. Seasonality. Figures 5(a) graphs the aver- or bonuses at the end of each quarter. Figure 5(b)
age revenues over the months for the regular sales shows the index of indirect sales revenue (ISR) for
Figure 4(b) Sales and Percentage Quota Achieved—Pre-Bonus Months
3 3
May sales/May quota
Feb sales/Feb quota
2 2
1 1
0 0
0 0.2 0.4 0.6 0.8 1.0 1.2 0 0.2 0.4 0.6 0.8 1.0 1.2
% quarterly quota sold by Jan % quarterly quota sold by Apr
3 3
Aug sales/Aug quota
Nov sales/Nov quota
2 2
1 1
0 0
0 0.2 0.4 0.6 0.8 1.0 1.2 0 0.2 0.4 0.6 0.8 1.0 1.2
% quarterly quota sold by July % annual quota sold by Oct
Figure 5(a) Revenues from the Regular Sales Force Table 5 Testing for Sales Substitution Across Quarters
300 Model 1 Model 2
∗∗∗
Other months of quarter 168087 101076∗∗∗
200 4340575 4340355
First month of quarter 147079∗∗∗ 101086∗∗∗
4340875 4340525
100 Other months of quarter × 91009∗∗∗ 56058∗∗∗
Previous month % distance to quota 450625 450905
First month of quarter × 2059 0015
0 Previous month % distance to quota 470025 460935
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
ISR (indirect revenue) index 38072∗∗∗
420265
Figure 5(b) ISR Index Salesperson fixed effects Yes Yes
5
Note. Standard deviations are shown in parentheses.
∗∗∗
4 p < 0001.
3 for, there is a gap in revenue that we interpret as

induced by effort. It is interesting that these gaps
2
are larger at the end of the quarter, suggesting the
1 value of bonuses in inducing effort. We empirically
0
estimate the multiple for the ISR index to control for
seasonality. We later show in §5.1 that the ISR mul-
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
tiple for our model is about 25, given that the aver-
Notes. The ISR index is on a base of 1, reflecting the indirect sales in January age of lagged annual quota is 1,639. We acknowledge
1999. Here, we have averaged the index across three years (1999–2001).
that accounts handled by the direct and indirect sales
Figure 5(c) Revenues from the Regular Sales Force and ISR Index force are likely systematically different, but our main-
Multiples tained assumption is that the seasonality multipliers
300 on sales are identical across the two types of accounts.
Revenues ISR * 30 We acknowledge that this assumption is a limitation
250 ISR * 40 ISR * 50 to fully control for seasonality.
200 2.3.3. Sales Substitution Across Quarters. One
possibility is that salespeople giving up at the end
150
of the quarter may be doing so to increase the odds
100 of meeting quotas in subsequent quarters by simply
not booking the sales in the current quarter. Alter-
50
natively, a salesperson who is meeting the quarterly
0 quota may simply shift sales to the next quarter to
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec increase the odds of meeting next quarter’s quota.
In either case, we should see a positive relationship
the pure commission indirect sales force.9 Because between the first month of the quarter and the previ-
the ISR index is not contaminated by bonuses, we ous quarter’s percentage cumulative performance to
use it to control for seasonality and isolate the effect quota (%QQ or %AQ). However, in Table 5, we find
of bonuses on salesperson effort and revenues. To that the coefficient of First month of quarter × Previous
build intuition for how the ISR index can help con- month (%) distance to quota is statistically insignificant
trol for seasonality and isolate effort, see Figure 5(c), and very small in magnitude.10 We therefore abstract
which graphs the average revenues of the direct sales away from the challenges of modeling sales substi-
force and multiples of the ISR index. At a multiple of tution in our application. We acknowledge that our
around 50, the ISR virtually mimics the average rev- test may have limited power in identifying timing
enues, making the revenues from the commissioned behavior; future research should explore more seri-
and bonus sales force close to identical. This sug- ously the issue of sales timing and the attendant mod-
gests that bonuses are not effective in inducing addi- eling challenges.
tional effort. When the ISR index has a multiple of 30
or 40, even after the overall seasonality is accounted 10
Managers of the focal firm stated that they monitor sales agents
extensively to prevent such time shifting of sales bookings. Also,
9
For the indirect sales force, the firm only provided us with an index customers play a major role in determining when sales occur in
of revenues. The ISR index is set to a base of 1 for January 1999. this environment.
3. Model policy function is different for each month. Further-

Based on the model-free evidence, we build a more, we allow a time-invariant salesperson-specific
dynamic model of sales force response to the quota- variable, tenure with the firm (), to moderate the
based compensation scheme. The timing of the model level of effort.
is as follows: Note that unlike the demand-shifter function f 4 · 5,
1. At the beginning of each year, the firm chooses which is common across all salespeople, the effort
an annual compensation plan. function will vary across salespeople. Specifically, we
2. Each month, agents observe their current state allow for salespeople to belong to one of multiple dis-
and exert effort in a dynamically optimal manner. crete segments; hence these effort functions will be
3. An idiosyncratic sales shock is realized; the estimated at the segment level. We estimate the effort
shock plus the agent’s effort determines the agent’s function nonparametrically using Chebyshev polyno-
realized sales for the period. The agent receives com- mials of the variables described above.
pensation. 3.2. Compensation Plan
4. The realized sales of the current period affect the The compensation plan has three components: (i) the
agent’s state of the next period. Steps 2 and 3 are monthly salary wit ; (ii) the end-of quarter bonus Biqt
repeated each month until the end of the year, and received for achieving the corresponding quarterly
Steps 1–3 are repeated each year. quota Qiqt , and the end-of-year bonus Biyt received for
We describe the model in five parts: (i) the com- achieving the corresponding annual quota Qiyt ; and
pensation plan, (ii) the sales agent’s utility function, (iii) a commission rate, rit , per dollar’s worth of sales
(iii) state transitions, (iv) the effort as a function of and an overachievement commission rate, rit0 , given
state variables, and (v) the optimal effort choice by at the end of the year for sales over and above the
the sales agent. annual quota for each individual i at time t. We rep-
resent the compensation plan for a salesperson i by
3.1. Sales Response Model
the vector it = 8wit 1 Qiqt 1 Qiyt 1 Biqt 1 Biyt 1 rit 1 rit0 9.
We model the sales revenue function (Sit ) for sales-
person i at time t in two parts: (1) a base level of 3.3. Salesperson’s Per-Period Utility
sales independent of effort, parameterized by demand In each period t, salesperson i receives a positive util-
shifters (zDit ), and (2) sales induced as a result of effort ity of wealth Wit earned based on realized sales and
(eit ), parameterized by effort shifters that include ter- a disutility C4eit 3 i 5 from exerting effort eit . Thus the
ritory and salesperson characteristics (zEit ): utility function is defined as
Sit = f 4zDit 5 + eit 4zEit 5 + it 1 (1) U 4eit 1Sit 3i 1i 1i 5
= E6W 4Sit 3i 57−i var6W 4Sit 3i 57−C4eit 3i 51
where it is an additive sales revenue shock not antic-
ipated by the salesperson when choosing effort. where i and i are the risk aversion and disutility
As discussed previously, the market potential varies parameters, respectively, for salesperson i.11
across territories and across time. To account for the Given the sales levels and the compensation plan,
cross-sectional variation in market potential, we use the wealth for individual i, Wit , can be computed;
the annual quota from the previous year (AQi1 y−1 ). To Wit arises from four components: the per-period
account for seasonality of demand across months, we salary component wit , the lump-sum bonus compo-
use the ISR index (ISRt ). We also include an interac- nent Bit , the commission component Cit , and the
tion between the two variables to allow for season- overachievement commission component OCit . The
ality to have a different impact on differently sized detailed expressions of wealth is as follows:
territories.
Wit = wit + Bit + Cit + OCit 1
For effort shifters in eit , we use the following
sit 4eit 4zEit 51 zDit 3 i 5 + it

variables: given that effort is a function of demand Bit = Iqt I zi1t + > 1 Bqt
shifters, we include both AQi1 y−1 and ISRt in zEit . As Qiqt
discussed in the motivation, the salesperson’s state
sit 4eit 4zEit 51 zDit 3 i 5 + it

with respect to achieving her quota will have an + Iyt I zi2t + > 1 Byt 1
impact on the effort she expends. We therefore use Qiyt
the cumulative percentage of quarterly and annual
11
quotas completed till time t (%QQit 1 %AQit ) as vari- In the case of the constant absolute risk-aversion utility function
ables that affect effort. In addition, we use the time (exponential utility function) with normal errors and a linear com-
pensation plan, this functional form represents the certainty equiv-
(month type) within quarterly and annual quota peri- alent utility of the agent. Here, we consider the utility function to
ods to allow for different effort levels at different tem- be a second-order approximation to a general concave utility func-
poral distances to bonus payments; thus, the effort tion with a constant level of risk aversion.
Cit = 4sit 4eit 4zEit 51 zDit 3 i 5 + it 5rit 1 Whereas the first two state variables evolve stochas-
tically, conditional on the effort levels and revenue
s 4e 4zE 51 zD 3 5 + it

OCit = Iyt I zi2t + it it it it i >1 shocks of the previous periods, the latter two evolve
Qiyt in a purely deterministic manner. Other state vari-
· zi2t Qiyt + sit 4eit 4zEit 51 zDit 3 i 5 + it − Qiyt r 0it 1 ables would include time-varying demand shifters,

the ISR index, and territory characteristics for which
where zi1t and zi2t are the percentage of quarterly and we use the previous year’s annual quota. We use
annual quotas completed, respectively, by salesper- tenure with the focal firm () as an individual state
son i until time t; Iqt and Iyt are indicators for whether variable that impacts effort. These state variables are
time t is a quarterly or annual bonus period. collected in a state vector,
In our empirical analysis, we use a quadratic func-
zEit = zi1t 1 zi2t 1 zi3t 1 zi4t 1 ISt 1 AQi4y−15 1 i 0

tional form for the disutility function; specifically,
C4e3 i 5 = i e2 . Thus the set of structural parameters
3.5. Optimal Choice of Effort
of the salesperson’s utility function that need to be
Given the parameters of the compensation scheme ,
estimated is i = 4i 1 i 5.
as well as the state variables and their transitions,
each sales agent would choose an effort level con-
3.4. State Variables
ditional on her states to maximize the discounted
As discussed previously, the nonlinearity of the com- stream of expected future utility flow. Alternatively, if
pensation scheme with quotas and bonuses intro- this value function is below the reservation wage, the
duces dynamics into the sales agent’s behavior salesperson may choose to leave the firm.
because there is an additional trade-off between the The stream of utility flow, under the optimal effort
disutility of effort today and a higher probability of a policy function, conditional on staying at the firm, can
lump-sum bonus and overachievement commissions be represented by a value function:
tomorrow. To incorporate the dynamics of the model,
we consider the following key state variables: the per- V 4z3 1 ì5
centage of annual quota completed, the percentage n h
of quarterly quota completed, month type within the = max U 4e1 z3 1 5 + E max U 4e0 1 z0 3 1 5
e e0
quarterly quota period, and month type within the h iio
00 00
annual quota period (time of the year). These state + E max 00
U 4e 1 z 3 1 5 + · · · 1
e
variables evolve as follows:
1. Percentage of quarterly quota completed (%QQ) where ì = 41 5 is the set of primitives or structural
parameters of the underlying utility function and

0 if t is the start of the quarterly the discount factor . The expectation of the value

 quota period1 function is taken with respect to both the present and
zi1t = Si4t−15 future sales shocks.
zi14t−15 + otherwise3


Qiqt
4. Estimation
2. Percentage of annual quota completed (%AQ)
Traditionally, the nested fixed-point algorithm (NFXP)
 developed by Rust (1987) is used to estimate dynamic
0 if t is the start of the annual

 models. However, NFXP estimators are computa-
 quota period1
zi2t = Si4t−15 tionally burdensome because one has to solve the
zi24t−15 +

 otherwise3 dynamic program numerically over each guess of the
Qiyt parameter space for every iteration. The two-step esti-
mation first introduced by Hotz and Miller (1993) and
3. Month type within quarterly quota period
extended by Bajari et al. (2007) can serve to reduce the
 computational burden. In this approach, the model
1 if t is the start of the quarterly
estimation proceeds in two steps. In the first step, we
zi3t = quota period1
estimate the conditional choice probabilities of choos-
zi34t−15 + 1 otherwise3

ing a certain action as a flexible nonparametric func-
tion of state variables. Then, in the second step, these
4. Month type within annual quota period
conditional choice probabilities are used to estimate
 the structural parameters of the sales agent’s utility
1 if t is the start of the annual
function.
zi4t = quota period1
Until recently, it was believed that the accurate
zi44t−15 + 1 otherwise0

estimation of conditional choice probabilities for an
agent is impractical when there is unobserved het- The lagged annual quota takes into account general
erogeneity. Arcidiacono and Miller (2011) propose an territory characteristics that are likely to be generated
expectation-maximization algorithm-based approach with limited effort (i.e., market size). The revenues
to accommodate unobserved heterogeneity in the first from the indirect sales force capture market seasonal-
step of the two-step estimation procedure. We provide ity, independent of the nonlinear nature of the com-
one of the first applications of this approach illustrat- pensation plan. We assume that the revenue shocks
ing the empirical validity of the approach in practical (it ) come from an independent and identically dis-
applications. We now discuss the details of the two- tributed normal distribution.14
step estimation procedure. If it were possible to estimate the sales response and
effort response functions at the level of each individ-
4.1. Step 1: Estimating CCPs ual, then one could simply obtain the individual-level
In this step, we need to estimate a flexible nonpara- parameters of the effort and sales policy functions by
metric mapping between observable states and the maximizing the log-likelihood of the sample as
salesperson’s actions; this requires a nonparametric T L
model of the monthly effort function, ei 4zEit 5, that links D E
X X
ä̂i = argmax log Li Sit −f 4zit 3i 5− l 4zit 5il 1
effort and states in Equation (1). We model the effort äi t=1 l=1
function nonparametrically as a combination of basis (3)
functions of the state variables. Thus the nonparamet- where the vector äi = 8i 1 i 1 i 9 contains the set of
ric effort function is parameters of the sales response and effort policy
functions and the distribution of sales shocks, where
L
eit =
X
l 4zEit 5il 1 (2) 1 2
Li 45 = e−41/254/i 5 0 (4)
l=1 i 4251/2
where the lth basis function is l 4zEit 5. In this applica- We accommodate unobserved heterogeneity by
tion, the lth basis function is the lth-order Chebyshev allowing for discrete segments. Assume that salesper-
polynomial.12 son i belongs to one of K segments, k ∈ 811 0 0 0 1 K9
with segment probabilities qi = 8qi1 1 0 0 0 1 qiK 9. Let the
From Equations (1) and (2), we have the following
population probability of being in segment k be k .
sales response function to estimate:
Let L4Sit zit 1 k3 äk 5 be the likelihood of individual i’s
L sales being Sit at time t, conditional on the observ-
Sit = f 4zDit 5 + l 4zEit 5il + it 0
X
ables zit and the unobservable segment k, given seg-
l=1 ment parameters äk . Then the likelihood of observ-
ing sales history Si over the time period t = 11 0 0 0 1 T ,
For zDit , which is a subset of zEit (hereafter referred to
given the observable history zi and the unobservable
as zit 5, we use two variables: (i) the lagged annual
segment k, is given by
quota for salesperson i and (ii) the ISR index. We
T
use the direct linear effect of these variables to con-

Y
trol for cross-sectional variations of territory charac- Lk 4Si zi 3 äk 1 k 5 = k Likt 1 (5)
t=1
teristics and temporal variations in monthly seasonal-
ity. The interaction effects of these variables with the where Likt = L4Sit zit 1 k3 äk 5. As noted earlier, we
other state variables go into the polynomial function assume the distribution of the revenue shocks to be
in (2).13 normally distributed and hence use the normal likeli-
hood for Equation (5) as in Equation (4). The param-
12
eter äk = 8k 1 k 1 k 9 is the vector of segment-level
Chebyshev polynomials are a sequence of orthogonal polyno-
parameters of the sales response and effort policy
mials, defined by the recurrence relation T0 4x5 = 1, T1 4x5 = x, and
Tn+1 4x5 = 2xTn 4x5 − Tn−1 4x5. By using the orthogonal Chebyshev functions where each k is the parameter that indexes
polynomials to approximate the policy function, rather than the effort policy for segment k, and k is the param-
approximate the policy function with the standard 1, x, x2 poly- eter for the distribution of the revenue shocks for
nomials, we avoid multicollinearity issues. In our approximation segment k.
of the effort policy function (reported in Table 6(b)), we use up to By summing over all of the unobserved states
third-order Chebyshev polynomials; i.e., T0 4x5 = 1, T1 4x5 = x, T2 4x5 =
2x2 − 1, and T3 4x5 = 4x3 − 3x. For more details on Chebyshev poly-
k ∈ 811 0 0 0 1 K9, we obtain the overall likelihood of
nomials, we refer the reader to Judd (1998). individual i:
13 K
Separating sales effects stemming from seasonality or territorial X
characteristics and effort is a difficult challenge given the unobserv- L4Si zi 3 ä1 5 = Lk 4Si zi 3 äk 1 k 51
ability of effort. Our maintained assumption that the main effect of k=1
zDit isolates the territory and seasonal effects from the effort policy
14
function is valid if the firm sets quotas in proportion to territory It is noteworthy that Mirrlees (1999) shows that the first-best can
characteristics and seasonality (zDit ). This remains an important issue be approximated with normal errors in the sales response function
to explore in future research. in a one-shot model setting if unbounded punishments are feasible.
and hence the log-likelihood over the N sample of Thus, this procedure gives us the sales revenue func-
individuals becomes tion Ŝ4 · 5 and effort policy function ê4 · 5 for each seg-
N N K T ment. We use these segment-level policies to obtain the
structural parameters of each segment. A caveat with
X X X Y
log L4Si zi 3 ä1 5 = log k Likt 0 (6)
i=1 i=1 k=1 t=1 the two-step estimation procedure is that the first-
stage policy function estimates can be biased (and thus
Directly maximizing the log-likelihood in (6) is
generate bias in the structural parameters) if the state
computationally infeasible because the function is
variables in the policy function are correlated with the
not additively separable. So we use the approach of
first-stage errors.
Arcidiacono and Jones (2003) and Arcidiacono and
Miller (2011) to iteratively maximize the expected log- 4.2. Step 2: Estimating Structural Parameters
likelihood in Equation (7): The key idea of the two-step estimation is that in
N X
K X
T
the first stage we observe the agent’s optimal actions.
Using these observed optimal actions, we are able
X
qik log L4Sit zit 1 k3 äk 51 (7)
i=1 k=1 t=1 to construct estimates of the value function, which
enables us to estimate the primitives of the model that
where qik is formally defined below as the probability rationalize these optimal actions.
that individual i is of segment type k given parameter Let the value function of a representative agent at
values ä = 8ä1 1 0 0 0 1 äK 91 where äk = 8k 1 k 1 k 9, and state z that follows an action profile e—conditional on
segment probabilities = 81 1 0 0 0 1 k 9, conditional on the compensation plan , the sales profile S, and the
all of the observed data of individual i: primitives of the utility function and discount param-
eters ì = 41 5—be represented as
Pr4k Si 1 zi 3 ä1 5 = qik 4Si 1 zi 3 ä1 5
Lk 4Si zi 3 äk 1 k 5 Vt 4z3 e3 1 S1 ì5
= 0 (8)
L4Si zi 3 ä1 5 T
X

=E D4t5U 4e4zt 51 zt 1 t 3 5 z0 = z3 1 S1 ì 1 (9)
The iterative process is as follows: We start with an t=0
initial guess of the parameters ä0 and 0 . A natural
candidate for such starting values would be to obtain where D4t5 = t is the discount factor, and the expec-
the parameters from a model without unobserved tation operator would be over the present and future
heterogeneity and slightly perturbing those values.15 sales shock t .
Given the parameters 8äm 1 m 9 from the mth iteration, Using the estimated sales and effort policy function
the update of the (m + 1)th iteration is as follows: and the distribution of the sales shocks in the first
(a) Compute qik
4m+15
using Equation (8) with äm stage, we are able to forward-simulate the actions of
and . m sales agents to obtain the estimate of the value func-
(b) Obtain ä4m+15 by maximizing (7) evaluated tion. The detailed simulation procedure is as follows:
4m+15 (a) From initial state of zt , calculate the optimal
at qik .
actions as e4zt 5.
(c) Update 4m+15 by taking the average over the
(b) Draw sales shock t from f 45.
sample such that
(c) Update state zt+1 using the realized sales
1 XN s4e4zt 55 + t .
4m+15 4m+15
k = q 0 (d) Repeat (a)–(c) until t = T .
N i=1 ik
By averaging the sum of the discounted stream
We iterate (a)–(c) till convergence. of utility flow over multiple simulated paths, we
For the basis functions in the effort policy, we use can get the estimate of the value function, Ṽ 4z3 e4z53
Chebyshev polynomials of state variables to approxi- 1 S1 ì5.16
mate effort, as described in footnote 12. From the esti- Let es 4z5 be any deviation policy from a set of
mation, we obtain the vector of parameters for the feasible policies that is not identical to the optimal
basis functions (), the vector of parameters for policy, and, by using the same simulation method
the sales policy (), and the parameters of the revenue proposed above, let the corresponding estimate of
shocks () for each segment k. We also obtain the pop- the value function be called the suboptimal value
ulation segment probabilities () for each segment. function, Ṽ 4z3 es 4z53 1 S1 ì5. Since e4z5 by definition
is the effort policy and thus at an optimum, then
15
any deviations from this policy rule would generate
We started the initial values from one-tenth of the standard error
from the parameter values obtained from a single-segment model.
16
The initial values of the segment probabilities were set equally For each segment, we drew 400 simulation draws over each
across segments. period and computed the value functions.
value functions of lesser or equal value to that of the simulate data with the estimated structural param-
optimal level. eters associated with each discount factor to com-
Let us define the difference in the two value func- pute the mean absolute percentage error (MAPE) per
tions as period with respect to the observed data. We choose
the final estimates as the discount factor and the asso-
Q4v3 1 S1 ì5 = V 4z3 e4z53 1 S1 ì5 ciated structural parameters with the lowest MAPE.20
− V 4z3 es 4z53 1 S1 ì51
4.3. Identification
where v ∈ V denotes a particular 8z1 e 4z59 combina- s There are a couple of major identification challenges.
tion.17 Then if e4z5 is the optimal policy, the function First, we do not observe effort. Hence the link
Q4v3 1 S1 ì5 would always have a value greater than between effort and sales cannot be identified nonpara-
or equal to zero. Thus our estimate of the underlying metrically. Second, in dynamic structural models, it
structural parameters ì would satisfy is usually impossible to identify discount factors sep-
Z arately from the utility function. Below, we discuss
2
ì̂ = arg min min8Q4v3 1 S1 ì51 09 dH 4v51 how we address these issues.
Realized sales are a function of demand shifters,
where H 4v5 is the distribution over the set V of in- effort, and additive sales shocks. Conditional on
equalities. Our empirical counterpart to Q4v3 1 S1 ì5 observed demand shifters and given multiple obser-
would be vations of sales at different states, we can separately
identify nonparametrically the density of sales shocks
Q̃4v3 1 Ŝ1 ì5 = Ṽ 4z3 ê4z53 1 Ŝ1 ì5 and a deterministic function of effort. We assume a
deterministic (but flexible) relationship between effort
− Ṽ 4z3 ês 4z53 1 Ŝ1 ì50 and observable states (%QQ and %AQ and demand
shifters) for each segment. Finally, because we do not
As a result, our estimates of the structural parameters
observe effort, we need a strictly monotonic paramet-
are obtained from minimizing the objective function
ric relationship between sales and effort. As we esti-
in Equation (10):18
mate a flexible relationship between observable states
NI and effort, we model the relationship between sales
1 X 2
min8Q̃4vj 3 1 Ŝ1 ì51 09 0 (10) and effort to be linear.
NI j=1 We rely on the variation in states across agents and
within agents for identification. The average coeffi-
The above procedure is performed for each
cient of variation of states %AQ within agents (i.e.,
segment with the segment-specific effort policies %AQ %AQ AQ AQ
CV = 1/N Ni=1 i /i , where i and i are
P
obtained in Step 1. This allows us to estimate the
the mean and standard deviation, respectively, of the
structural parameters for each segment.19 In practice,
distance to annual quotas over three years for agent i)
since the objective function in Equation (10) is rela-
over the three years is 0.55. This allows us to separate
tively flat with respect to changes in the discount fac-
effort from the individual fixed effects. The average
tor, it is difficult to pin down the discount parameter
coefficients of variation across agents and months on
with traditional gradient-based optimization. In our
%AQ and %QQ are 0.63 and 0.91, respectively.21 The
specification, we first estimate the structural param-
variation in states across agents helps separate agent
eters by minimizing the objective function in Equa-
effort from the monthly fixed effects.
tion (10) over a grid of discount factors. Second, we
The discount factor is not identified separately from
the utility function in standard dynamic structural
17
As indicated in Bajari et al. (2007), there are multiple ways to models because, typically, there are no variables that
draw these suboptimal policy rules. Although the method of select-
do not affect contemporaneous utility, only future
ing a particular perturbation will have implications for efficiency,
the only requirement necessary for consistency is that the distri-
bution of these perturbations has sufficient support to yield iden- 20
The statistical properties of the grid-search estimator based on
tification. We chose to draw a deviation policy from a normal MAPE for discount factors are not known. Hence, we computed
distribution with mean zero and a quarter of the variance from the the standard errors using subsample bootstrap methods, where
revenue shock distribution; i.e., es 4z5 = e4z5 + . we subsampled the data and estimated the model multiple times
18
We drew 200 deviation strategies to construct the objective func- to construct a bootstrap distribution (Efron 1979). In her analysis
tion, and hence NI =200. of entry and exit in the washing machine industry in the United
19
In addition, we used a second set of moment inequalities to reflect States, Shen (2013) also uses grid search to estimate the discount
the participation constraint that employees continued to work at a factor of firms; she uses parametric bootstrapping to estimate the
firm because they at least obtained a reservation value (normalized standard error.
21
to zero); i.e., min8V 4z3 e4z53 1 S1 ì51 09. It turns out these inequali- There is substantial variation in the across-agent CV of %AQ and
ties are nonbinding and do not affect our estimates. %QQ across months.
utility (Rust 1994, Magnac and Thesmar 2002). In the 5.1. First-Stage Estimates
absence of such an exclusion restriction, this implies The parameter estimates for the demand shifters in
that if an agent exerts low effort in a period, it is not the sales response function is reported in Table 6(a).
possible to distinguish whether this is due to high We find that lagged annual quota and the interaction
disutility for effort or because she discounts future term between it and indirect sales revenue are statis-
utilities very heavily.22 tically significant. Thus larger markets tend to have a
Two aspects of our setting allow us to identify util- bigger sales multiplier independent of effort in high-
ity functions separately from the discount factor. First, demand periods.
we have a finite-horizon setting, where at the end We estimate segment-level effort policy functions
of the year, the quotas are reset and all agents start by estimating the nonparametric relationship between
with a fresh quota for the following year. This means sales and state variables through Chebyshev poly-
that every December, the agent faces a static optimiza- nomials of the state variables. We estimated up to
tion problem, conditional on the sales agent’s state fourth-order Chebyshev polynomials with an alterna-
(%AQ). Utility parameters are well identified for a tive number of segments and chose the best-fitting
model based on the Bayesian information criterion.
static model, and hence the agent’s choice in the last
The best fit was for the model with up to third-order
period should allow us to nonparametrically identify
Chebyshev polynomials, allowing for three segments.
the agent’s utility function. Given this, a variation in
The estimates of the best-fitting polynomial function
sales (that is monotonically linked to effort) in the
and the standard deviations of the revenue shocks
last period and variations in wealth should help iden-
for each segment are reported in Tables 6(b) and 6(c).
tify the effort disutility and risk-aversion coefficient
As the coefficients associated with the Chebyshev
within the utility function. polynomials have no intuitive meaning, for intuition,
Second, the bonus setting generates exclusion we show graphs of the effort policy function for the
restrictions between the current and future utility; i.e., three segments as a function of the percentage annual
we have instruments in nonbonus periods that do not quota (%AQ) and percentage quarterly quota (%QQ)
affect current utility, only future utility. As we demon- for select months in Figure 6(a). Both %AQ and %QQ
strated with reduced-form evidence earlier, that an are normalized across sales agents, such that 1 implies
agent’s performance in November is related to his meeting quota and 0.9 indicates 10% below quota and
proximity to the annual bonus given in December 1.1 indicates 10% above quota.
indicates forward-looking behavior. We also collected Table 7 shows the share of the three segments
other evidence of how agents respond to quarterly or and their descriptive characteristics. Segment 2 is the
annual quotas even though they do not affect current largest with a share of 47%; Segments 1 and 3 have
payoffs. This allows us to estimate the discount factor. shares of 32% and 21%, respectively. The average
Beyond these conceptual arguments, we report in tenure with the firm is not very different across seg-
the appendix the results of a simulation study that ments, at approximately 12 years. Segment 3 has the
demonstrate that the structural parameters of the util- highest annual quotas, followed by Segments 2 and 1.
ity function and discount factor can be identified It is worth noting that Segments 2 and 3, with larger
using our empirical strategy. quotas, achieve their quota targets more often than
Segment 1, which has trouble meeting its quota.
Figure 6(a) shows that Segment 3 exerts the most
5. Results effort and is the most productive segment, and Seg-
We first report the first-stage estimates of the demand ment 1 exerts the least effort and is the least pro-
shifters and effort policy function for the sales ductive segment. This is consistent with the allocated
response model; then we report estimates of struc- quotas and percentage of time quotas are achieved in
tural parameters of sales agents’ utility functions from Table 7. We also see a positive relationship between
the second-stage estimation. We then perform several exerted effort and %AQ for all months shown. As for
counterfactual simulations to address the substantive Table 6(a) Parameter Estimates—Sales Response
questions we seek to answer.
Lagged annual quota −00022∗∗∗
4000075
22
Another reason why an agent might exert low effort is that she Indirect sales −00955
may have incorrect expectations about the transition density of 4603075
future states; i.e., she might be very pessimistic about future good Indirect sales × Lagged annual quota 00023∗∗∗
states. Similar to other dynamic structural modeling papers, we 4000035
assume rational expectations for the transition densities of states.
In this case, this translates into a rational expectations assumption Note. Standard deviations are shown in parentheses.
∗∗∗
on sales shocks. p < 0001.
Table 6(b) Parameter Estimates—Effort Policy Function
Variable Segment 1 Segment 2 Segment 3 Variable Segment 1 Segment 2 Segment 3

∗∗∗ ∗∗ ∗ ∗∗∗
0 −138025 −94017 −214082 1 4z2 51 4z3 5 2074 196033 11011025∗∗∗
4370815 4450725 41160775 4370735 4490875 41380315
1 4z1 5 59015 163096 308024 1 4z2 5 1 4z2 5 × 1 4z3 5 −56061 −168063∗∗∗ −101005
4780255 41010855 42480035 4530895 4640365 41540005
2 4z1 5 −32043 −42048 −201092∗∗∗ 1 4z4 5 −7044∗∗∗ −3085∗∗ −2053
4220065 4280765 4770665 410165 410665 430745
3 4z1 5 6053 11025∗ 27030∗ 1 4z1 51 4z4 5 14048 −2068 −13097
440685 450995 4150225 4150945 4210585 4580565
1 4z2 5 259022∗∗∗ 35067 −685074∗∗∗ 1 4z2 51 4z4 5 15054∗∗∗ 26000∗∗∗ 108023∗∗∗
4630825 4790015 41950495 440415 450275 4130485
2 4z2 5 −146079∗∗∗ −85054∗∗ −61003 1 4z2 5 1 4z2 5 × 1 4z4 5 −15044 −17099 54090
4300455 4340515 4780185 4130565 4140825 4370565
3 4z2 5 28072∗∗∗ 12055∗∗ 8090 1 4z1 5 1 4z3 5 × 1 4z4 5 −4060 8041 10096
450895 460055 4130375 450545 470545 4200435
1 4z1 51 4z2 5 74025 151011 −349073 1 4z2 5 1 4z3 5 × 1 4z4 5 −0025 −20080∗∗∗ −102095∗∗∗
41670365 41990475 44900915 430875 450175 4140635
1 4z1 52 4z2 5 1047 82006∗∗∗ −58059 1 4 5 −0016 8017∗∗∗ 15058∗∗
4350905 4310815 4680075 410385 420125 460815
2 4z1 51 4z2 5 16051 −24049 49078 2 4 5 0003 −0024∗∗∗ −0041
4200835 4240105 4530175 400055 400075 400265
1 4z1 51 (ISR) −37024∗∗∗ −65019∗∗∗ −57022 3 4 5 0000 0000∗∗∗ 0000
4120685 4150985 4400655 400005 400005 400005
1 4z2 51 (ISR) −13025∗∗∗ −0061 −38038∗∗∗ 1 4z1 51 4AQ y −1 5 −0004∗∗ −0003∗∗ −0007∗∗∗
430355 430915 4100275 400025 400025 400035
1 4z2 5 1 4z2 5 × 1 (ISR) 37042∗ 47050∗∗ 2084 1 4z2 51 4AQ y −1 5 −0001 −0006∗∗∗ −0002∗
4210485 4230675 4530515 400015 400015 400015
1 4z3 5 −0016 −10000∗∗ −43033∗∗∗ 1 4z2 5 1 4z2 5 × 1 (AQy −1 5 0004 0006∗ 0003
430395 450105 4110925 400045 400035 400045
1 4z1 51 4z3 5 54044∗∗ 81066∗∗ 113002
4260975 4370015 4970945
Note. Standard deviations are shown in parentheses.

∗
p < 001; ∗∗ p < 0005; ∗∗∗ p < 0001.
%QQ, we see an increasing but concave relationship tially increase effort with experience, but this tapers
in March, implying that once a salesperson is way off with time. This is probably because in the early
above the quarterly quota, she starts to gradually years of their careers, they want to work hard not only
slow down. Given that the average states in March for monetary payments from increased wages but also
for each segment were 0.55, 0.58, and 0.62, respec- for other intangible incentives, such as promotions or
tively, not a lot of salespeople are in a position where transfers to better job titles. However, after a certain
they can slow down. Effort in December does not number of years, these intangibles do not matter as
fall off even if the salesperson has already reached or much, and the effort levels tend to taper off. It is note-
exceeded her quota (%AQ > 1), likely because of the worthy that Segment 1, the lowest-productivity seg-
overachievement commissions in preventing sales- ment, does not gain in productivity from experience.
people from lowering effort after achieving quotas.
Our results are consistent with Steenburgh (2008), Table 7 Descriptive Characteristics of Segments
who finds that salespeople “give up” when far away
Segment 1 Segment 2 Segment 3
from achieving their quota, such as for all segments
in our case, but they do not slow down much once Share 0032 0047 0021
their quota is reached. Tenurea 11021 12034 11060
Figure 6(b) shows the effect of tenure on effort for Achieve quarterly quota: Q1 0046 0054 0057
Achieve quarterly quota: Q2 0038 0055 0062
all segments. Salespeople in Segments 2 and 3 ini-
Achieve quarterly quota: Q3 0030 0049 0053
Achieve annual quota 0030 0057 0065
Table 6(c) Revenue Shock Distribution Average annual quotab 11199008 11623011 21349010
Average December revenueb 127030 272053 564064
Segment 1 Segment 2 Segment 3
a
Tenure is measured in years.
Sigma 80.59 141.84 271.73 b
Average quotas and revenues are indicated in U.S. dollars (’000).
Figure 6(a) Effort Policy by Segment as a Function of Percentage of Quota
July December March

200 500 250 Seg 1
400 Seg 2
150 200
Seg 3
300
Effort
Effort
150
Effort
100
200 100
50
100 50
– – –
0.25
0.33
0.40
0.48
0.55
0.63
0.70
0.78
0.85
0.93
1.00
0.50
0.58
0.66
0.74
0.82
0.90
0.98
1.06
1.14
1.22
1.30
0.20
0.26
0.32
0.38
0.44
0.50
0.56
0.62
0.68
0.74
0.80
%AQ %AQ %QQ
5.2. Second-Stage Structural Parameter Estimates earned by the sales force, risk aversion is not a serious
The first column of Table 8 shows the estimates of concern. The estimated model fits the observed sales
the dynamic model with exponential discounting. We revenue data reasonably well with a MAPE of 9.4%.
find the discount factor to be 0.9 and highly sig-
nificant. Frederick et al. (2002) have a comprehensive 5.3. Assessing the Value of a
summary of the estimated discount factors from pre- Dynamic Structural Model
vious studies. The summary shows that the estimated How important is it to model the dynamics of sales-
discount factors vary extensively, ranging from as low person behavior? In a static model, any effort would
as a mere 0.02 to no discounting at all, with a discount be attributed to the current payoff, not accounting
factor of 1. For purely monetary values, the estimated for the large future bonuses. This will downward
discount factor seems rather low. But as Frederick bias the salesperson’s disutility parameters and over-
et al. point out, for behavioral aspects such as pain— state the effects of compensation on productivity. The
and in our case, effort—the discount factors tend to second column of Table 8 reports the estimates of
be low. Hence, our estimate appears to be reasonable. the myopic model, where the discount factor is set
The parameters for disutility of effort are negative to zero. As expected, the disutility parameters are
and significant for all three segments. Their relative smaller in magnitude relative to the forward-looking
magnitudes are consistent with the effort policy func- model for all segments. For Segment 3, the downward
tions estimated in the first stage. Segment 3, which bias is as much as 28%. The myopic model also has a
produces the greatest sales on average, has the lowest
disutility for effort. Segment 1, which has the low-
est sales, has the greatest disutility. The risk-aversion Table 8 Structural Parameters
coefficients for all segments are insignificant, show- Dynamic Myopic
ing no direct evidence of risk aversion by the sales
∗∗∗
agents. This may be because in the range of incomes Discount factor 0090
400045
Segment 1
Figure 6(b) The Effect of Tenure on Effort Disutility −00375∗∗ −00284∗∗
4001895 4001145
100
Risk aversion −000001 −000001
Seg 1 40000205 40000035
Seg 2
80
Segment 2
Seg 3 Disutility −00145∗∗∗ −00111∗∗∗
4000385 4000325
60 Risk aversion −000001 −000001
40000055 40000055
Effort
Segment 3
40 Disutility −00077∗∗∗ −00056∗∗∗
4000165 4000105
Risk aversion 000000 000000
20 40000025 40000045
MAPE 00094 00167
– Note. Standard errors are shown in parentheses.
1 2 3 4 5 6 7 8 9 10 11 12 13 ∗∗
p < 0005; ∗∗∗ p < 0001.
Tenure (years)
Figure 7 Simulated Revenue and Effort—Static vs. Dynamic
(a) Revenue (b) Effort

250 100
200 80
USD (K)
USD (K)
150 60
100 40
Dynamic
50 20
Myopic
0 0
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
poorer fit: a MAPE of 16.7% relative to MAPE of 9.4% agents from each segment (a total of 75 salespeople)
for the dynamic model. and simulated 1,000 simulation paths for each of them
We next compare the revenue and effort predictions to obtain the productivity of each segment. Then, we
between the dynamic and myopic models. To isolate weighted each segment based on the estimated seg-
the effects of forward-looking behavior, we simulate ment sizes reported in Table 7.
based on the disutility estimates from the dynamic
model but set the discount parameter to zero for the 5.4.1. Value of Nonlinear Incentive Components.
myopic model. Figure 7 compares the predicted rev- We compare changes in revenues and profits when
enues and effort of the myopic and dynamic models. the firm moves from the current compensation plan to
The myopic agent has systematically lower revenues a pure commission-only plan. We consider two cases:
because she does not take into account the effect of (1) where the commission rate is the same as the cur-
future bonuses and overachievement commission in rent commission rate and (2) where a higher commis-
current effort. In contrast, the forward-looking agent sion rate is such that total compensation is exactly
anticipates that in an uncertain environment, there is equal to the current compensation. We find that the
a chance of bad shocks later that may prevent her revenues are about 17.9% greater with the current
from meeting the quota, so she prepares for such a compensation plan compared with a pure commission
rainy day by working harder early on so that she is plan. Not surprisingly, given the lower incentives in
within striking distance of her quota even if a bad the absence of quotas, bonuses, and overachievement
sales shock occurs. commissions, all segments suffer from substantially
The effort graph in Figure 7 enables us to iso- poorer performance, as shown in Table 10. Even after
late out the sales revenue cyclicality and focus on adjusting commission rates to be higher to make total
the differences in effort across dynamic and myopic compensation identical to the current plan, revenues
agents. The myopic salesperson concentrates much are 2.4% higher with the current compensation plan,
more effort in the bonus period, but the forward- suggesting that the nonlinear plan shifts salespeople
looking salesperson smooths effort over time, given into high-power areas of incentives. To understand
the uncertainty in future demand shocks. The effort the role of different components, we next investigate
peaks in the bonus periods are not as pronounced counterfactuals based on individual components one
for the dynamic agent. The observed effort smooth- at a time.
ing is similar to consumption smoothing by forward-
looking consumers facing uncertain incomes in the Table 9 Impact of Alternative Bonus Plans on Sales Revenues
development economics literature. Counterfactual Change in revenues (%)
5.4. Counterfactual Simulations Only pure commissions −1709

We now perform a series of counterfactual simula- Only pure commissions −204
(adjusted to equal payout with bonus)
tions that address the two sets of substantive ques-
No bonus (only commissions + −800
tions we wish to answer. First, we address the issue
overachievement commission)
of how valuable different components of the com- No bonus (commissions adjusted to −206
pensation plan are. The overall change in revenues equal payout with bonus)
under the alternative conditions is reported in Table 9 No overachievement commissions −1007
and the effect by segment in Table 10. Second, we Cumulative annual quota replaced with −107
compare the role of bonus frequency—how quarterly quarterly quota
and annual bonuses affect performance. In running Annual bonus split into quarterly and −005
annual bonus
the counterfactual simulation studies, given a regime
Remove quarterly bonus −309
change, we randomly chose 25 representative sales
Table 10 Impact of Alternative Bonus Plans on Sales Revenues by Segment
Percentage of decrease from

different components Segment 1 (%) Segment 2 (%) Segment 3 (%)
Pure commission 1005 1904 1903

Without overachievement 204 1008 1405
Without quarterly bonus 600 404 202
5.4.2. Value of Overachievement Compensation. quota by replacing the cumulative annual quota with
When overachievement commissions are eliminated, just a fourth-quarter quota. That is, we set the fourth-
not surprisingly, overall revenues drop by 10.7%. quarter quota as the annual quota minus the sum of
Even accounting for the additional commission costs, three previous quarters’ quotas 4QQ4 = AQ − QQ1 −
profits are lower by about 2% (assuming a gross mar- QQ2 − QQ3 5 and split the total bonus payments
gin of 33%). (annual bonus + quarterly bonuses) equally across all
To gain insight into how overachievement commis- four quarters.23 Furthermore, to isolate the effect of
sions impact sales, Figure 8(a) compares the effort bonuses, we also remove the overachievement com-
level of sales agents who eventually meet or do not mission for reaching the annual quota in both sce-
meet the annual quota. For those who met the annual narios. Overall, revenues drop by 12.0% when the
quota, their effort level does not decline even when cumulative annual quota and overachievement com-
close to the quota because of the overachievement missions are removed. This decrease is greater than
commission. In contrast, those who did not meet the the 10.7% we obtained when we just dropped the
annual quota decrease effort toward the end of the
overachievement commission. Thus the cumulative
year because they are unlikely to meet the quota,
annual quota induces sales agents to exert greater
and therefore the overachievement commission has
effort and raise revenues by 1.3%.
no impact on their earnings. Thus the overachieve-
How does the cumulative annual quota work?
ment commission provides incentives for the most
On the negative side, the cumulative annual quota
productive salespeople even if they have already met
their quota (or are likely to meet it). Table 10 indi- reduces the incentive of salespeople who have not
cates, not surprisingly, that overachievement commis- had success in the first three quarters to continue to
sions have the greatest impact on Segment 3, the most put in effort in the fourth quarter. But on the positive
productive segment. Revenues drop by about 15% side, the cumulative annual quota allows salespeople
for this segment, whereas the effect on the least pro- who reach their quarterly quota in earlier quarters to
ductive segment is substantially smaller, at 2%. Over- continue to extend themselves in order to get them
all, we conclude that overachievement commission closer to the annual target—making this additional
rates motivate high-performing salespeople to con- effort early on helps sales agents buffer against nega-
tinue exerting high levels of effort when they are close tive shocks in future quarters and allows them to be
to and exceed the quota, but these rates have a limited within striking distance of the annual quota. On bal-
impact on low-performing salespeople. ance, the positive effect of the latter outweighs the
5.4.3. Value of a Cumulative Annual Quota. negative effect of the former.
Next we investigate the effect of cumulative annual We also consider the case where we split the cur-
rent annual bonus into a fourth-quarter bonus (the
same amount as other quarterly bonuses) and an
Figure 8(a) Effort: Overachievement Commission
annual bonus so that salespeople do not give up in
120
the last quarter when they are far away from quota.
Met AQ
110 Specifically, we split the $4,000 annual bonus into
Did not meet AQ
100 a $1,500 fourth-quarter bonus and a $2,500 annual
bonus. Although this did increase the effort in the
90
USD (K)
last quarter, it reduced revenues overall because sales-

80 people did not put in as much effort earlier in the
70 year to be within striking distance of the annual quota
60
because it is not as large. Total revenues dropped
by 0.5%.
50
40
23
Specifically, the quarterly bonus is set to be (annual bonus + 3 ×
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
quarterly bonus)/4; i.e., (4,000 + 11500 × 35/4 = $2,125.

Figure 8(b) Effort: Quarterly Quotas experimental work in behavioral psychology (Heath
100 et al. 1999). The basic idea is that achieving short-term
Current plan goals makes achieving long-term goals more feasi-
90
Without quarterly bonus ble. Our analysis shows that the short-term goals are
more valuable to the least productive segment; i.e., in
USD (K)
80
education terms, it implies that weaker students gain
70
more by periodic testing relative to stronger students
who would study regardless of exams.
60 5.4.5. Discussion. Overall, our results provide
empirically grounded substantive insight on the role
50
of various elements of nonlinear compensation to
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
improve productivity. The numerical example in §2.2
Overall, these results show that the cumulative shows that quotas and bonuses serve as important
annual quota has a significant impact on salespeo- goals for the average performers, inducing them to
ple by inducing agents to work harder even when increase their effort. Our empirical analysis shows
they reach their quarterly targets, again motivating support for the role of quotas and bonuses as motiva-
the high-performing salespeople who reach quarterly tional goals and stretch incentives, respectively. How-
ever, it is important to recognize that other elements
quotas to continue to remain productive and bring
of the compensation plan are critical in improving
additional sales.
the performance of both high performers and low
5.4.4. Quota-Bonus Frequency. We next investi- performance. First, in this setting, managers recog-
gate the value of quarterly bonuses relative to annual nize that quota updating based on the salesperson’s
bonuses. Figure 8(b) shows the comparison of effort own performance in the previous year can induce
between the current plan and when quarterly bonuses ratcheting effects; such effects can reduce productiv-
are eliminated and only the annual bonus is left. ity among the highest-performing salespeople, who
Effort drops consistently across the year when there do not book sales much higher than their quota so as
are no quarterly quotas. Overall revenues fall by 4%. to avoid higher quotas the following year. By updat-
Even in December, when the annual bonus is on the ing quotas based not on an individual’s performance
table, revenue falls by 1% and effort falls by 2%. but on the performance of a larger group of sales-
Thus annual bonus and overachievement commis- people, managers can minimize ratcheting effects.
sions have less of an impact on year-end performance In fact, we found empirically little statistical evidence
without quarterly bonuses. Why? of effort shading as a result of ratcheting effects. Sec-
The quarterly bonus induces sales agents to work ond, the high performers who are most likely to reach
harder in a given quarter. But it also helps them their quotas will reduce effort upon reaching them if
achieve the annual quota by helping them stay on there are no additional incentives offered. The firm
track of their annual goal. Without a quarterly bonus, used overachievement commissions and cumulative
sales agents do not have much incentive to work annual quotas to induce the high-performing sales
hard early on. This lack of incentive leads them to agents to continue to put in effort and excel in their
be farther away from the annual quota by Decem- sales performance. These incentives, along with the
ber. Annual bonuses and overachievement commis- avoidance of ratcheting effects, enabled the firm to
sion have little impact on effort as sales agents are keep its best sales agents motivated to continue per-
more likely to give up meeting the annual quota. forming at a high level. Finally, quarterly bonuses
The impact of quarterly bonuses also differs across served as pacers to help keep the least productive
the three segments of consumers. Table 10 indicates segment motivated so that it can be within striking
that quarterly bonuses have a relatively minimal distance of its long-term goals. By providing empir-
impact on Segment 3, the most productive segment, ically grounded insight into the differential motiva-
but they almost triple the impact of Segment 3 on tional roles of different incentive components of a
Segment 1. In effect, quarterly bonuses are needed as nonlinear compensation plan for high-, average-, and
pacers for the less productive salespeople than for the low-performing sales agents, this paper contributes
most productive salespeople. substantively to literature on the role of nonlineari-
To the best of our knowledge, there has been no ties in quota-bonus plans and explains its widespread
analysis to date on what the appropriate frequency of popularity in practice (Joseph and Kalwani 1998).
quota and bonuses is. There has been some descrip-
tive work in the education literature on how frequent 6. Conclusion
testing affects academic performance (for an exten- Personal selling is a primary marketing mix tool for
sive survey, see Bangert-Drowns et al. 1991) and some most B2B firms to generate sales, yet there is little
research on how the compensation plan motivates selection issues. One possibility is to use a longer
the sales force and affects performance. This paper panel of salespeople’s performance that includes attri-
develops and estimates a dynamic structural model tion information. If there were variation in contracts
of sales force response to a compensation plan with that affected employee retention, that could also help
various components: salary, commissions, lump-sum address this problem. More work on scenarios with
bonuses for achieving quotas, and different commis- richer contracts needs to be done. For example, one
sion rates beyond achieving quotas. Our analysis could study peer effects on sales performance and
helps us assess the impact of (1) different components selection effects when firms shift from individual- to
of compensation and (2) the differential importance team-based compensation (Chan et al. 2014).
of periodic bonuses on the performance of different In summary, this paper provides a rigorous frame-
types of salespeople. work to empirically understand how the sales force
We find that the quota-bonus scheme used by this responds to a very rich compensation structure involv-
firm increases the sales force’s performance by serving many components of compensation: salaries,
ing as an intermediary goal and pushing employ- commissions, quotas, and bonuses at quarterly and
ees to meet targets. Features such as overachieve- annual frequencies. Our analysis helps obtain a num-
ment compensation reduce the problems associated ber of useful substantive insights. Nevertheless, the
with sales agents slacking off when they get close to issues raised above provide an interesting agenda for
achieving their quota. Furthermore, quarterly bonuses future work.
serve as a continuous evaluation scheme to keep
sales agents within striking distance of their annual Acknowledgments
quotas. In the absence of quarterly bonuses, failure This paper is the first essay of the first author’s doctoral
in the early periods to meet targets cause agents to dissertation at Yale University. The authors thank the editor,
fall behind more often than in the presence of quar- associate editor, two anonymous reviewers, Lanier Benkard,
Adam Copeland, David Godes, Ahmed Khwaja, Subrata
terly bonuses. Thus, the quarterly bonus serves as a
Sen, as well as the seminar participants at Carnegie Mellon
valuable subgoal that helps the sales force stay on University, Cornell University, Emory University, the Geor-
track in achieving its overall goal; such incentives gia Institute of Technology, Harvard University, the Hong
are especially valuable to low performers. In contrast, Kong University of Science and Technology, the Indian
overachievement commissions increase performance Institute of Management, the Indian School of Business,
among the highest performers. New York University, Northwestern University, Pennsylva-
In this empirical application, we introduce two nia State University, Stanford University, the University of
important methodological innovations to the market- Arizona, the University of British Columbia, the University
ing literature. First, to the best of our knowledge, of California at Berkeley, the University of Chicago, the Uni-
we provide the first empirical implementation of the versity of Houston, the University of Maryland, the Univer-
Arcidiacono and Miller (2011) approach to accommo- sity of North Carolina, the University of Rochester, Wash-
date unobserved heterogeneity within a two-step esti- ington University, Western University, and Yale University,
in addition to the conference participants at the 2009 UT
mation framework. Second, we demonstrate that dis-
Dallas FORMS Conference, the 2010 Choice Conference in
count factors can be estimated in naturally occurring Key Largo, the 2010 Sales Conference in Kansas, the 2012
field data using appropriate exclusion restrictions— Harvard Thought Leadership on the Sales Profession Con-
and we overturn the conventional wisdom that esti- ference, and the 2012 INFORMS Marketing Science Confer-
mation of discount factors requires augmentation ence for their comments and suggestions. The authors also
with survey data. thank ISBM for providing financial support.
We now discuss the limitations of our paper, which
Appendix. Simulation Study to Identify
provide promising avenues for future research. First, Discount Factors
effort tends to be multidimensional, and one possibil- Let realized sales be a function of effort, a function of
ity is that quotas and bonuses force people to focus demand shifter zt , and a normal random sales shock it
on the effort that leads to final sales in bonus peri- such that
ods, whereas agents may focus on the earlier stages
of the selling process in nonbonus periods. Such a yit = h4zt 5 + eit + it 1 it ∼ N401 2 50
multidimensional effort cannot be identified merely
To illustrate identification, we consider a simulation sce-
from sales data. We hope data from customer rela-
nario with seasonality and quarterly quotas. The demand
tionship management databases that track customer shifter zt represents the seasonality index comparable to
stages through the selling process can help shed light the ISR index in our empirical setting. For the purpose
on this issue. We believe this is an exciting area for of the simulation, we assume the function h4 · 5 is lin-
future research. ear in the seasonality index; i.e., h4zt 5 = zt . Let the
Second, compensation contracts can serve to select salesperson’s utility be given as uit = −deit2 + Wit , where
the right type of salespeople. We do not address Wit = BI8t=31 61 91 129 I8sit +yit >Q9 is the wealth earned from bonus
Table A.1 Simulation Results Chevalier J, Goolsbee A (2009) Are durable goods consumers
forward-looking? Evidence from college textbooks. Quart. J.
True values Econom. 124(4):1853–1884.
No. of
individuals d = −001 = 009 = 10 Copeland A, Monnet C (2009) The welfare effects of incentive
schemes. Rev. Econom. Stud. 76(1):96–113.
50 −00134 00892 100282 Darmon R (1997) Selecting appropriate sales quota plan structures
4001665 4002875 4003205 and quota-setting procedures. J. Personal Selling Sales Manage-
100 −00106 00899 100266 ment 17(1):1–16.
4000255 4000065 4002215 Dubé J-P, Hitsch G, Jindal P (2012) The joint identification of utility
and discount functions from stated choice data: An applica-
200 −00108 00899 100043
tion to durable goods adoption. Working paper, University of
4000185 4000035 4001505
Chicago, Chicago.
300 −00105 00899 90944 Efron B (1979) Bootstrap methods: Another look at the jackknife.
4000105 4000025 4001215 Ann. Statist. 7(1):1–26.
Note. Standard errors are in parentheses. Ferrall C, Shearer B (1999) Incentives and transactions costs within
the firm: Estimating an agency model using payroll records.
payment B for achieving quarterly quota Q. The cumulative Rev. Econom. Stud. 66(2):309–338.
sales state within quarter evolves as follows: Finger SR (2012) An empirical analysis of R&D competition in
the chemicals industry: Incorporating unobserved heterogene-
if t is the start of quota period1 ity into a dynanic game estimator. Working paper, University
sit = 0
si4t−15 + yit otherwise0 of South Carolina, Columbia.
Frederick S, Loewenstein G, O’Donoghue T (2002) Time discount-
We report the results of the simulation with seasonality ing and time preference: A critical review. J. Econom. Literature
and quarterly quotas using three years of data. Let the vec- 40(2):351–401.
tor of the seasonality index for each month throughout the Heath C, Larrick RP, Wu G (1999) Goals as reference points. Cogni-
year be given by (21 31 51 11 21 31 31 41 71 21 31 5). We set the tive Psych. 38(1):79–109.
bonus at B = 60 and quarterly quota at Q = 30. Consistent Heide CP (1999) Dartnell’s 30th Sales Force Compensation Survey,
with the model in the paper, we estimate the first-stage esti- 30th ed. (Dartnell Corp, Durham, NC).
mates using our specification in Equation (1). Holmstrom B (1979) Moral hazard and observability. Bell J. Econom.
We varied the simulated number of individuals from 50 10(1):74–91.
to 300. The true values and the estimates and standard Holmstrom B, Milgrom P (1987) Aggregation and linearity in
errors for each simulation are reported in Table A.1. Even the provision of intertemporal incentives. Econometrica 55(2):
with 100 individuals (relative to our sample size of 350), we 303–328.
are able to recover the disutility parameter and the discount Hotz VJ, Miller RA (1993) Conditional choice probabilities and
the estimation of dynamic models. Rev. Econom. Stud. 60(3):
factor with reasonable precision, lending confidence to the
497–529.
identification arguments in §4.3.
Jiang R, Palmatier RW (2009) Structural estimation of a moral haz-
ard model: An application to business selling. Working paper,
References University of California, Davis, Davis.
Joseph K, Kalwani M (1998) The role of bonus pay in sales
Albers S, Mantrala MK, Sridhar S (2010) Personal selling elasticities:
force compensation plans. Indust. Marketing Management 27(2):
A meta-analysis. J. Marketing Res. 47(5):840–853.
147–159.
Arcidiacono P, Jones JB (2003) Finite mixture distributions, sequen-
Judd KL (1998) Numerical Methods in Economics (MIT Press, Cam-
tial likelihood, and the EM algorithm. Econometrica 71(3):
bridge, MA).
933–946.
Arcidiacono P, Miller RA (2011) Conditional choice probability esti- Kim SK (1997) Limited liability and bonus contracts. J. Econom.
mation of dynamic discrete choice models with unobserved Management Strategy 6(4):899–913.
heterogeneity. Econometrica 79(6):1823–1867. Lal R, Srinivasan V (1993) Compensation plans for single- and
Assmus G, Farley JU, Lehmann DR (1984) How advertising affects multi-product sales forces: An application of the Holmstrom-
sales: Meta-analysis of econometric results. J. Marketing Res. Milgrom model. Management Sci. 39(7):777–793.
21(1):65–74. Magnac T, Thesmar D (2002) Identifying dynamic discrete choice
Bajari P, Benkard CL, Levin J (2007) Estimating dynamic models of models. Econometrica 70(2):801–816.
imperfect competition. Econometrica 75(5):1331–1370. Mirrlees JA (1999) The theory of moral hazard and unobserv-
Bangert-Drowns RL, Kulik JA, Kulik C-LC (1991) Effects of frequent able behaviour: Part I. Rev. Econom. Stud. 66(1):3–21. [Mimeo,
classroom testing. J. Educational Res. 85(2):89–99. Nuffield College, Oxford, UK, 1975.]
Basu AK, Lal R, Srinivasan V, Staelin R (1985) Salesforce compen- Misra S, Nair H (2011) A structural model of sales-force compen-
sation plans: An agency theoretic perspective. Marketing Sci. sation dynamics: Estimation and field implementation. Quant.
4(4):267–291. Marketing Econom. 9(3):211–257.
Beauchamp A (2012) Regulation, imperfect competition, and the Oyer P (1998) Fiscal year ends and nonlinear incentive contracts:
U.S. abortion market. Working paper, Boston College, Chestnut The effect on business seasonality. Quart. J. Econom. 113(1):
Hill, MA. 149–185.
Bijmolt THA, Van Heerde HJ, Pieters RGM (2005) New empirical Oyer P (2000) A theory of sales quotas with limited liability and
generalizations on the determinants of price elasticity. J. Mar- rent sharing. J. Labor Econom. 18(3):405–426.
keting Res. 42(2):141–156. Paarsch HJ, Shearer B (2000) Piece rates, fixed wages, and incen-
Chan TY, Li J, Pierce L (2014) Compensation and peer effects in tive effects: Statistical evidence from payroll records. Internat.
competing sales teams. Management Sci. Forthcoming. Econom. Rev. 41(1):59–92.
Park E-S (1995) Incentive contracting under limited liability. J. Steenburgh T (2008) Effort or timing: The effect of lump-sum
Econom. Management Strategy 4(3):477–490. bonuses. Quant. Marketing Econom. 6(3):235–256.
Raju JS, Srinivasan V (1996) Quota-based compensation plans
Sudhir K (2001) Structural analysis of manufacturer pricing in the
for multi-territory heterogeneous salesforces. Management Sci.
presence of strategic retailer. Marketing Sci. 20(3):244–264.
42(10):1454–1462.
Rao RC (1990) Compensating heterogeneous sales-forces: Some Training (2008) Quota-based compensation on the rise. (Octo-
explicit solutions. Marketing Sci. 9(4):319–341. ber 15) http://www.trainingmag.com/article/quota-based
Rust J (1987) Optimal replacement of GMC bus engines: An empir- -compensation-rise.
ical model of Harold Zurcher. Econometrica 55(5):999–1033.
Yao S, Mela CF, Chiang J, Chen Y (2012) Determining con-
Rust J (1994) Structural estimation of Markov decision processes. sumers’ discount rates with field studies. J. Marketing Res.
Engle RE, McFadden D, eds. Handbook of Econometrics, Vol. 4
49(6): 822–841.
(Elsevier-North Holland, Amsterdam), 3082–3139.
Shen Q (2013) A dynamic model of entry and exit in a grow- Zoltners AA, Sinha P, Lorimer SE (2008) Sales force effectiveness:
ing industry. Working paper, University of Pennsylvania, A framework for researchers and practitioners. J. Personal Sell-
Philadelphia. ing Sales Management 28(2):115–131.

MKSC 2013 0815

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

MKSC 2013 0815

Uploaded by

Copyright:

Available Formats

This article was downloaded by: [134.7.34.

232] On: 29 October 2018, At: 08:43

Do Bonuses Enhance Sales Productivity? A Dynamic

To cite this article:

Full terms and conditions of use: http://pubsonline.informs.org/page/terms-and-conditions

Copyright © 2014, INFORMS

Please scroll down for article—it is on subsequent pages

Do Bonuses Enhance Sales Productivity?

1. Introduction Relative sales force expenditure elasticity is useful

Figure 1 Types of Incentive Compensation Schemes

Plan A: Pure commission Plan B: Pure bonus Plan C: Commission at quota

Plan D: Commission with Plan E: Commission + Bonus Plan F: Commission + Bonus

Table 1 Firm’s Compensation Plan

Type Description Payment period

Table 2 Descriptive Statistics 2.2. Numerical Example

Figure 2(a) How Quotas and Bonus Serve as Stretch Goals

d = 1, r = 10, B = 0, e* = 5 d = 1, r = 10, B = 30, e* = 10 d = 1, r = 20, B = 0, e* = 10

Figure 2(b) Effort as a Function of Distance to Quotas

d = 2, r = 10, B = 30, d = 2, r = 10, B = 30 d = 2, r = 10, B = 30

–150 –100 –100

%AQ < 005 %AQ > 005

0.6 Note. Standard deviations are shown in

Two key elements stand out from Figure 4(a). First,

Figure 4(a) Sales and Percentage Quota Achieved—Bonus Months

Jun sales/Jun quota

Dec sales/Dec quota

Figure 4(b) Sales and Percentage Quota Achieved—Pre-Bonus Months

Nov sales/Nov quota

3 for, there is a gap in revenue that we interpret as

3. Model policy function is different for each month. Further-

Table 6(b) Parameter Estimates—Effort Policy Function

Variable Segment 1 Segment 2 Segment 3 Variable Segment 1 Segment 2 Segment 3

Note. Standard deviations are shown in parentheses.

Figure 6(a) Effort Policy by Segment as a Function of Percentage of Quota

July December March

Figure 7 Simulated Revenue and Effort—Static vs. Dynamic

(a) Revenue (b) Effort

5.4. Counterfactual Simulations Only pure commissions −1709

Table 10 Impact of Alternative Bonus Plans on Sales Revenues by Segment

Percentage of decrease from

Pure commission 1005 1904 1903

last quarter, it reduced revenues overall because sales-

quarterly bonus)/4; i.e., (4,000 + 11500 × 35/4 = $2,125.

You might also like