You are on page 1of 19

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/318357621

Bayesian analysis of working capital management on corporate profitability:


evidence from India

Article  in  Journal of Economic Studies · July 2017


DOI: 10.1108/JES-11-2015-0207

CITATIONS READS

16 740

3 authors, including:

Arvind Shrivastava
Reserve Bank of India
16 PUBLICATIONS   33 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Corporate distress and Default Prediction, Trade Credit, Working Capital Management, Econometrics and Time Series Modelling View project

Structural Breaks-Bayesian Approach View project

All content following this page was uploaded by Arvind Shrivastava on 28 August 2017.

The user has requested enhancement of the downloaded file.


Journal of Economic Studies
Bayesian analysis of working capital management on corporate profitability:
evidence from India
Arvind Shrivastava, Nitin Kumar, Purnendu Kumar,
Article information:
To cite this document:
Arvind Shrivastava, Nitin Kumar, Purnendu Kumar, (2017) "Bayesian analysis of working capital
management on corporate profitability: evidence from India", Journal of Economic Studies, Vol. 44
Issue: 4, pp.568-584, https://doi.org/10.1108/JES-11-2015-0207
Permanent link to this document:
Downloaded by Reserve Bank of India At 23:18 27 August 2017 (PT)

https://doi.org/10.1108/JES-11-2015-0207
Downloaded on: 27 August 2017, At: 23:18 (PT)
References: this document contains references to 21 other documents.
To copy this document: permissions@emeraldinsight.com
The fulltext of this document has been downloaded 40 times since 2017*
Users who downloaded this article also downloaded:
(2017),"Working capital management and firm profitability: a meta-analysis", Qualitative
Research in Financial Markets, Vol. 9 Iss 1 pp. 34-47 <a href="https://doi.org/10.1108/
QRFM-06-2016-0018">https://doi.org/10.1108/QRFM-06-2016-0018</a>
(2017),"Working capital management and firms’ profitability: evidence from emerging Asian
countries", South Asian Journal of Business Studies, Vol. 6 Iss 1 pp. 80-97 <a href="https://
doi.org/10.1108/SAJBS-09-2015-0060">https://doi.org/10.1108/SAJBS-09-2015-0060</a>

Access to this document was granted through an Emerald subscription provided by emerald-
srm:386026 []
For Authors
If you would like to write for this, or any other Emerald publication, then please use our Emerald
for Authors service information about how to choose which publication to write for and submission
guidelines are available for all. Please visit www.emeraldinsight.com/authors for more information.
About Emerald www.emeraldinsight.com
Emerald is a global publisher linking research and practice to the benefit of society. The company
manages a portfolio of more than 290 journals and over 2,350 books and book series volumes, as
well as providing an extensive range of online products and additional customer resources and
services.
Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the
Committee on Publication Ethics (COPE) and also works with Portico and the LOCKSS initiative for
digital archive preservation.

*Related content and download information correct at time of download.


The current issue and full text archive of this journal is available on Emerald Insight at:
www.emeraldinsight.com/0144-3585.htm

JES
44,4 Bayesian analysis of working
capital management on corporate
profitability: evidence from India
568 Arvind Shrivastava
Department of Statistics and Information Management, Reserve Bank of India,
Received 12 November 2015
Revised 20 June 2016 Mumbai, India
Accepted 25 November 2016
Nitin Kumar
Reserve Bank of India, Mumbai, India, and
Purnendu Kumar
Department of Statistics and Information Management, Reserve Bank of India,
Downloaded by Reserve Bank of India At 23:18 27 August 2017 (PT)

Mumbai, India

Abstract
Purpose – Decisions pertaining to working capital management have pivotal role for firms’ short-term
financial decisions. The purpose of this paper is to examine impact of working capital on profitability for
Indian corporate entities.
Design/methodology/approach – Both classical panel analysis and Bayesian techniques have been
employed that provides opportunity not only to perform comparative analysis but also allows flexibility in
prior distribution assumptions.
Findings – It is found that longer cash conversion period has detrimental influence on profitability. Financial
soundness indicators are playing significant role in determining firm profitability. Larger firms seem to be
more profitable and significant as per Bayesian approach. Bayesian approach has led to considerable gain in
estimation fit.
Practical implications – Observing the highly skewed distribution of dependent variable, Multivariate
Student t-distribution has been considered along with normal distribution to model stochastic term.
Accordingly, Bayesian methodology is applied.
Originality/value – Analysis of working capital for firms has been performed in Indian context. Application
of Bayesian methodology is performed on balanced panel spanning from 2003 to 2012. As per author’s
knowledge, this is the first study which applies Bayesian approach employing panel data for the analysis of
working capital management for Indian firms.
Keywords Panel data, Working capital, Bayesian econometrics
Paper type Research paper

1. Introduction
Each firm, irrespective of size, sector and social objectives requires working capital for its
routine business activities. Although, corporate finance literature has focused considerably
on relationship between long term finance like investment, capital structure, dividends on
company valuation and performance (Levine et al., 2000; Levine, 2005), management of
working capital that portrays short term financing needs of company is also vital parameter
for firms’ policy decisions. Working capital management essentially is administration of
current assets and current liabilities. Working capital has crucial consequence on available
liquidity and profitability of a firm. Effective working capital management leads to decline
in cash conversion cycle (CCC), which is time span between disbursement and collection of

JEL Classification — C11, C23, G3


Journal of Economic Studies The authors are grateful to anonymous referees for useful comments. The study also benefitted
Vol. 44 No. 4, 2017
pp. 568-584 from inputs received in conference held at the Central University of Rajasthan, August 24-25, 2015.
© Emerald Publishing Limited The views expressed in this paper are exclusively of the authors and need not necessarily belong to the
0144-3585
DOI 10.1108/JES-11-2015-0207 organization to which they belong.
cash, improvement in profitability and company valuation for shareholders (Shin and Working
Soenen, 1998; Eljelly, 2004). capital
Working capital is defined as operational current assets minus current liabilities. management
It usually equates to cash, receivable accounts and inventory minus payable accounts and
delayed accounts. An optimal level of working capital is very desirable for smooth routine
operations (Harris, 2005). In case working capital management is not given due
considerations, firms can fail and face bankruptcy (Filbeck and Krueger, 2005). A high level 569
of current assets may reduce risk of liquidity by renting or leasing plant and machinery,
whereas, similar policy cannot be followed for components of working capital.
Strict collections policies and limited sales credits to clients may reduce period of
accounts receivable and improve cash inflows on one hand but may lead to foregone sales
translating to reduced profits. Also, fewer inventories run into risk of stock-outs. On the
other hand, liberal collection policies and longer credit implies less cash in hand leading to
unavailability of resources for other activities. However, estimating working capital needs of
Downloaded by Reserve Bank of India At 23:18 27 August 2017 (PT)

an entity at a specific time is a daunting task as it varies across nature of business,


production cycle, inventory availability among others. An understanding of role and drivers
of working capital can help in reducing risk and improving performance.
The existing study examines the relationship between firm profitability
and working capital management for Indian firms for the period 2003 to 2012.
Classical panel data analysis is performed to study impact of working capital on firm
profitability. However, in view of distributional limitation of classical approach, Bayesian
methodology is also applied that allows for exact inference, avoiding need to rely on
asymptotic distributions that can lead to severe small sample impediments. As per
frequentist approach, standard panel data analysis techniques of fixed or random effects
modeling exist that are based on assumption of Gaussian distribution. The work
contributes by applying Bayesian panel data model that allows flexible specification of
parameters. Multivariate Student t-non-Gaussian distribution is utilized that is able to
accommodate fat tails allowing for outliers and asymmetries in underlying data.
Application of Bayesian approach has gained prominence due to its flexibility
and accuracy by taking care of limitations of traditional techniques. There have
been numerous studies applying Bayesian econometrics to panel data that enables us to
exploit the large cross-section of firms to obtain more precise parameter estimates
(Koop et al., 2007). Consequently, the study also enables to perform a comparative analysis
between two econometric methodologies of regression. It focuses on non-financial and
non-government Indian firms to assess the impact of working capital management on
their performance in a panel framework. Unambiguously, it is observed that longer cash
conversion period has detrimental impact on firm profitability.
The paper proceeds as follows. Section 2 provides brief review of literature on working
capital management. Section 3 elaborates on data and variables employed in the analysis
followed by explanation of Bayesian methodology employed in Section 4. The results of
empirical analysis are presented in Section 5 with broad summary of overall results
discussed in Section 6.

2. Review of literature
Analysis of working capital management has been performed by researchers in numerous
ways. The basic interest lies in its impact on profitability or firm valuation. However, choice
of variables for optimal inventory level, accounts receivables, trade credit policies has varied
from case to case. Impact of working capital management on profitability for joint stock
companies of Saudi Arabia was studied by Eljelly (2004). An inverse relationship was
established between the two with working capital being captured by CCC. Both CCC and
firm size were found to be more pronounced at industry level.
JES Similarly, employing listed firms at Athens Stock Exchange, Lazaridis and Tryfonidis
44,4 (2006) examined relationship between profitability and working capital management.
Gross operating profit was used as the measure of profitability being strongly related to
CCC. The results indicated that individual components of CCC may be optimized by
management for profit maximization.
An alternative measure of working capital management had been employed by Shin and
570 Soenen (1998) who analyzed a sample of US firms spanning 1975 to 1994 using Net Trading
Cycle (NTC) as comprehensive measure of working capital management. NTC is essentially
CCC[1] expressed as a percentage of sales. Both correlation and regression analysis
were applied that clearly revealed significant negative relationship between NTC
and profitability.
Deloof (2003) analyzed sample of 1,009 Belgian firms for period 1992-1996 with number
of days accounts receivable, inventories and accounts payable used as measures of trade
credit and inventory policies. The CCC was utilized to capture working capital management.
Downloaded by Reserve Bank of India At 23:18 27 August 2017 (PT)

It was found that managers can increase corporate profitability by reducing number of
day’s accounts receivable and inventories.
In one of few studies for India, efficiency of working capital management of Indian
cement industry was performed by Ghosh and Maji (2004). Employing accounting ratios
from 1993 to 2002, efficiency index of working capital management was constructed that
was decomposed into performance and utilization ratios. On applying regression analysis
an improvement of working capital efficiency of cement industry was observed. Kumar and
Shrivastava (2013) discussed importance of trade credits which is part of working capital on
performance of company in terms of profitability using firm level data spanning from 2001
to 2011 in Indian context.
Hierarchical Bayesian panel data modeling was performed by Bauer et al. (2008)
to assess market β dynamics. Specification of hierarchical priors enabled capturing
cross-sectional heterogeneity in market β’s without need to estimate large number of
parameters. Bayesian modeling improved accuracy of β forecasts. Also, it was shown that
methodology delivered superior firm-specific βs controlling cross-sectional heterogeneity,
which is ignored in traditional approach leading to flawed estimates.
Mitchell et al. (2011) applied Bayesian model averaging approach to identify economic
and policy drivers of international migration to UK from 14 different source regions over
1980-2007. Bayesian approach was found to better control for heterogeneity and identify
in a robust manner determinants of migration. In similar vein, Moral-Benito (2012)
employed Bayesian model averaging to address issue of model uncertainty for cross
country growth analysis of 73 countries from 1960 to 2000. Results were found to be
robust to different prior assumptions.
Bayesian panel probit analysis had been performed by Amisano and Giorgetti (2013)
to study entry into pharmaceutical submarkets. Bayesian approach helped to
eschew reliance on asymptotic properties of classical estimators, choose flexible prior
and control for heterogeneity. The data set covered sales in seven different countries
for period 1987-1998. It was found that global and submarket size measures had
different effects and that sunk costs often had positive effects on entry due to commitment
mechanism.
Testing economic growth convergence of G-7 nations employing longitudinal data was
carried out by Meligkotsidou et al. (2012). To overcome challenges of cross-sectional
correlated effects and initial income conditions Bayesian framework was utilized. The paper
provided clear cut evidence of growth convergence and also showed that ignoring covariate
effects tends to support economic divergence hypothesis. Results demonstrated that
Bayesian approach can successfully distinguish stationary from non-stationary
specifications of AR (1) panel data model allowing for covariates.
3. Data and variables Working
The private corporate sector in India contributes around one-third in gross capital formation. capital
It comprises of more than 800 thousands active private companies providing employment to management
more than ten million people in 2013 with consistent growth over years.
The paper utilizes information of non-government and non-financial public limited
companies collated from their annual reports/balance sheets as obtained from Reserve Bank
of India. The state-owned and financial sector has been excluded from analysis due to their 571
varied firm objectives and separate regulatory structure. Finally, our database is a balanced
panel of 1,172 firms over period from 2003 to 2012, i.e. ten years.
Our dependent variable comprises of Gross operating profitability (GOPR), which is sum
of earnings before interest and tax normalized by total assets. Amongst prominent
explanatory variables are CCC. CCC is a standard measure utilized in literature to account
for working capital[2]. It is a metric that expresses length of time it takes for a concern to
convert resource inputs into cash flows. This measure illustrates how quickly a company
Downloaded by Reserve Bank of India At 23:18 27 August 2017 (PT)

can convert its products into cash through sales. The shorter the cycle, the less time capital
is tied up in the business process, and thus better for company’s available resources:

CCC ¼ I TD þACPAPP (1)

here ITD is defined as average period to convert input materials to final furnished products
that are afterwards sold. It shows how effectively inventory is managed by comparing cost
of goods sold with average inventory for a period. It is calculated as:

Inventory  365
I TD ¼
Cost of goods sold
ACP is average time required for transforming company’s receivables to cash. It measures
how effectively a company extends credit and collects debts. It is an activity ratio reflecting
asset usage efficiency of firm. It is calculated as:

Receivable accounts  365


ACP ¼
Sales
APP represents average period taken by a firm for making payments to its creditors. It is
derived as:

Payable accounts  365


APP ¼
Cost of goods sold
The average Net Trade Cycle (NTC) period has also been calculated (ratio of difference
between trade debtors and trade creditors to annual sales multiplied by 365) which
measures average number of days company grants net financing as a consequence of its
business transactions.
Financial soundness of a firm plays pivotal role in determining performance and
profitability of a firm. Consequently, indicators reflecting financial soundness taken in
model are firms’ leverage ratio (LR), i.e. debt to total assets ratio, ratio of current assets to
total assets (CATA), current liabilities to total assets ratio (CLTA), liquidity ratio, i.e. ratio of
current assets to current liabilities (CACL). CACL measures ability of a firm to pay its debts
in near future.
Certain control variables have been included to disentangle confounding effects amongst
variables. Size of firm is taken as natural logarithm of sales normalized by Wholesale Price
Index (WPI) and denoted as SIZE.
JES 3.1 Panel data model and methodology
44,4 As per structure of our database, observations of individual companies may not be
independent across varied years making OLS methodology inappropriate. So, standard
pooled data estimation methodology, i.e. fixed/random effects model has been applied.
Hausman test has been utilized to test appropriateness of fixed or random effects model.
The basic functional form of panel regression equation that is repeated observations on
572 same set of cross-section units is as follows:
ðGOPRÞit ¼ b0 þb1 ðCCC Þit þb2 ðSI Z E Þit þb3 ðLRÞit þb4 ðCACLÞit þb5 ðCLTAÞit þ
þb6 ðCATAÞit þb7 ðGW CTRÞit þai þEit (2)

ðGOPRÞit ¼ b0 þb1 ðCCC Þit þbsq ðCCC Þ2it þb2 ðSI Z E Þit þb3 ðLRÞit þ b4 ðCACLÞit
þb5 ðCLTAÞit þb6 ðCATAÞit þb7 ðGW CTRÞit þ ai þEit (3)
Downloaded by Reserve Bank of India At 23:18 27 August 2017 (PT)

Key variable CCC in Equation (2) is substituted by APP, NTC, ITD and ACP in alternative
regressions for comparative purpose. Equation (3) has an additional square term of CCC to
examine its non-linear relationship with dependent variable.

4. Bayesian estimation, prior distribution, posterior distribution and Monte


Carlo Markov Chain (MCMC) methods
Bayesian approach has gained prominence and acceptability not only in statistical
sciences but also in areas of economics, banking, finance, corporate finance, etc.
To introduce Bayesian approach in panel data model, let us define a panel regression
model as follows:
 
yit ¼ X it b þai þEit where Eit  N 0; t1 (4)

here yit represents value of endogenous variable for ith firm at tth period where i ¼ 1, 2,
…, N and t ¼ 1, 2, …, T. Xit stands for matrix of explanatory variables for firm i at time t.
β ¼ ( β1, β2, … βn)0 is vector of associated parameters. αi is treated as individual effect
and if it is uncorrelated with explanatory variables then there exists evidence for
random-effect model (REM) else fixed-effect model (FEM) is considered more appropriate
representation. ϵit is usual stochastic disturbance term following normal distribution with
mean 0 and variance σ2 or precision τ.
On stacking over time fixed-effect model can be re-written as:
 
yi ¼ X i b þai þEi where Ei  N 0; t1 I T (5)

here dependent variable yi is a vector of length T for ith firm. Xi is a T × K matrix and β is a
K × 1 vector of coefficients. The error term is of length T × 1 and is normal homoscedastic,
independent of Xi and no autocorrelation.
Let us arrange Equation (5) as fosllows:

2 3 2 32 3 2 3
y1 X1 1 0 : :
0 b E1
6y 7 6X 0 1 : : 7
0 76 a1 7 6 E 2 7
6 7 6
6 2 7 6 2 7
6 7 6 76 7 6 7
6: 7 ¼ 6: : : 1 : 07 6 : 7þ6 : 7
6 7 6 76 7 6 7
6: 7 6: : : : 1 07 6 7 6 7
4 5 4 54 : 5 4 : 5
yN XN 0 : : : 1 aN EN
Equivalently we can write above equation as follows: Working
  capital
y ¼ Z KþE; where E  N 0; t1 I NT (6)
management
where y is of size N × T and matrix of explanatory variables Z is of size NT × (K+N) and
coefficient vector Λ is of length K+N.
The parameter Λ is treated as random variable in Bayesian approach unlike fixed as in
non-Bayesian methodology where it is estimated using maximum likelihood estimation. 573
As per Bayesian methodology prior parameters Λ and τ of model M as given in (6) is
distributed as p(Λ) and p(τ), respectively. Parameter estimation in Bayesian
approach is based on observed data y and prior distribution p(Λ) and p(τ). Choice of prior
distributions is based on previous research or experience based on past data. The inference
of Λ and τ under given model is fully described by conditional distribution of Λ given model
M which is known as posterior distribution of parameters based on data. Application of
Bayesian technique involves derivation of joint posterior distribution p(Λ, τ/Y) and marginal
Downloaded by Reserve Bank of India At 23:18 27 August 2017 (PT)

posterior distributions p(Λ/Y) and p(τ/Y) that is main focus of interest and defined by Bayes
Law as:
 
  P Y =K; t  P ðK; tÞ
p K; t=Y ¼ (7)
P ðY Þ
here, p(Λ, τ/Y) denotes joint density function of posterior distribution of parameters
Λ and τ. Λ is coefficient vector, P(Y/Λ, τ) is likelihood function and distribution of
joint prior is denoted by P(Λ, τ). The selection of proper priors for parameters plays key
role in derivation of precise posterior distribution of parameters. It represents
distribution of possible parameter values. Mainly two kinds of prior distributions have
been discussed in literature. One is informative prior that is based on past data or from an
expert. Other one is non-information prior having no specific distribution.
Non-information prior or no prior are proportional to constant having large variance
and leads to minimum role in posterior inference of parameters based on sample data.
The Bayesian estimate of parameter is usually defined as mean or mode of parameter
posterior distribution.
The likelihood function of fixed-effect model under assumption of normal distribution is
given as:
     
L y=K; t pt 2 exp  t=2 ðyZ KÞ0 ðyZ KÞ
NT
(8)
The posterior distribution of parameters (Λ, τ) can be calculated using Equation (7) and
Bayes Laws as:
     
p K; t=y pp y=K; t p t=K pðKÞ (9)

and marginal posterior density of Λ and τ can be obtained as:


Z
     
p K=Y ppðKÞ p Y =K; t p t=K dt (10)

Z
     
p t=Y ppðtÞ p Y =K; t p K=t dK (11)

Inferences about parameters are based on posterior distribution. However, in applications,


posterior density usually gets complex and intractable due to higher order integrals.
JES Hence, derivation of marginal posterior densities does not come as nice closed form
44,4 solutions. Hence, efficient simulation methods of drawing observations are applied to
estimate marginal densities.
In such scenarios, numerical approximation of integrals is performed through
efficient computing processes. MCMC method is widely employed to derive marginal
densities. Finally, quartiles of posterior distribution can be estimated from simulated
574 observations. As an illustration, mean of sample points generated from posterior
distribution of parameter may represent Bayesian estimate of that parameter in
actual model. In case of known conditional posterior densities one may apply Gibbs
Sampling methodology and in case of unknown conditional posterior densities one can
utilize Metropolis-Hasting sampler scheme to obtain marginal posterior densities
of parameters.

4.1 MCMC sampling scheme


Downloaded by Reserve Bank of India At 23:18 27 August 2017 (PT)

The availability of efficient algorithms and powerful systems has made implementation of
Bayesian approach possible in wide variety of model estimations and hypothesis testing.
The MCMC sample scheme is widely used algorithm for Bayesian estimation.
Gibbs sampler is utilized to carry out MCMC sampling. The Gibbs sampler begins with
an initial set of starting values for parameters, under some general conditions. The sampling
distribution resulting from this sequence converges to target distribution as iteration goes
to infinite. Convergence can also be obtained from one chain, though often requiring a
considerably larger number of iterations. Once chain has stabilized, initial few values
referred to as “burn-in” sample are discarded to remove impurity which occurred in
sampling due to introduction of initial values of parameters in MCMC scheme. Summary
statistics, including posterior mean, mode, standard deviation and credibility intervals,
are calculated based on post burn-in iterations.
The random sample generating process from conditional posterior densities is as follows.
We start with initial values (Λ(0), τ(0)) and then draw a random sample from conditional
posterior distribution of Λ given τ ¼ τ(0) say Λ(1), then take Λ ¼ Λ(1) and draw a random
sample from conditional posterior density of τ given Λ ¼ Λ(1). Proceeding in this manner,
after q iterations, we obtain:
 
KðrÞ ; tðrÞ ; r ¼ 1; 2; . . .; q:

4.2 MCMC sequences


(1) Sample Lð1sÞ ppðL1 =Lð2s1Þ ; Lð3s1Þ ; . . . Lðqs1Þ ; tðs1Þ ; yÞ
(2) Sample Lð2sÞ ppðL2 =Lð1s1Þ ; Lð3s1Þ ; . . . Lðqs1Þ ; tðs1Þ ; yÞ
(3) Sample tðsÞ ppðt=Lð1s1Þ Lð2s1Þ ; Lð3s1Þ ; . . . Lðqs1Þ ; yÞ

where s ¼ (1, 2, … S) are Monte Carlo iterations.


The marginal posterior densities of Λ and τ are obtained as:
8
>
> Xq
 
>
> pn
ðK Þ ¼ 1
p KjtðsÞ
>
< q
s¼1
X (12)
>
>
q
 
>
> pn ðtÞ ¼ q1 p tjKðsÞ
>
: s¼1
Further, note that Bayes estimators (parameter estimates) of Λ and τ are given by: Working
8 capital
>
> X q
 
>
>K ^ ¼ 1 K tð s Þ management
>
< q
s¼1
X (13)
>
>
q
 
>
> ^
t ¼ 1
t K ðs Þ
>
: q
s¼1 575
As required in Bayesian framework, it is necessary to specify prior distribution. As a
starting point, prior is assumed to be normally distributed. Additionally, Multivariate
Student t-distributed prior has also been presumed. The Student t-distribution has been
chosen as it is established to be more exact and appropriate for longer than normal errors
providing robust statistical inference (Lange et al., 1989; Zellner and Ando, 2010).
Both random and fixed effects variants of model have been estimated. In case of fixed-effect
Downloaded by Reserve Bank of India At 23:18 27 August 2017 (PT)

model for normally distributed errors, prior distribution is specified in single stage. If priors
are independent, their combination leads to joint prior distribution of model parameters.
Combination of joint prior distribution with likelihood function leads to joint posterior
distribution. However, in case of random-effect model, prior distribution has hierarchical
structure, since it is specified in two stages. Comparison amongst competing models is
based on R2 criteria.
In case of fixed-effect model with errors normally distributed, prior distribution is
specified in single stage as:
 
b0  N m1 ; s21 ; m1 and s21 known for t ¼ 1; 2; . . .; 10

 
ai  N m2 ; s22 ; m2 and s22 known for i ¼ 1; 2; . . .; 1172

 
bkj  N m3 ; s23 ; m3 and s23 known for k ¼ 1; 2; . . .; 6 : j ¼ 1; 2

t1 ¼ s2  I Gða1 ; b1 Þ where a1 and b1 are known:


Parameter chains do not immediately converge to posterior distribution. As Bayesian
estimation is iterative process, checking for convergence is vital. The convergence is
checked utilizing trace plot that shows parameter history. One can also make more precise
inference about convergence by observing autocorrelation plots and marginal density plot
of model parameters. High autocorrelations imply that chains are slow to converge.
All diagnostics indicate that parameter chains have converged after 8,000 iterations. In our
empirical analysis we therefore run 40,000 iterations and discard first 8,000 iterations as
burn-in sample and select every 4th sample from iteration to overcome effect of
autocorrelation of random draws. The remaining draws are used to summarize posterior
density and to conduct Bayesian inference. Kernel density plots shape of (univariate)
marginal posterior distribution of parameter. It is an approximation of marginal posterior.
The kernel density plot and autocorrelation plot of parameters are shown in Figure A1.

5. Empirical analysis and results


Before discussing regression results, summary statistics is provided to obtain glimpse of
behavior of data set. Table I provides temporal trend of variables. Profitability indicator,
i.e. GOPR has shown improvement from 38 percent in 2003 to 41 percent in 2007. However,
after financial crisis it has broadly declined, dropping to 36 percent in 2012,
JES 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
44,4
GOPR 0.382 0.401 0.410 0.412 0.419 0.399 0.408 0.421 0.382 0.364
(0.269) (0.28) (0.277) (0.279) (0.284) (0.284) (0.293) (0.29) (0.284) (0.277)
ITD 97.455 94.276 90.538 90.476 86.983 88.152 83.760 87.575 87.117 82.747
(60.572) (60.707) (58.139) (58.3) (55.482) (56.298) (58.487) (59.123) (57.706) (56.497)
ACP 77.541 74.221 71.308 72.373 71.085 71.455 67.204 69.918 71.217 72.654
576 (49.624) (48.257) (47.066) (48.233) (46.926) (47.068) (47.81) (47.683) (48.199) (49.53)
APP 82.253 79.532 76.027 74.700 74.819 75.691 71.337 74.627 73.089 66.253
(55.162) (53.767) (51.354) (50.88) (50.858) (50.911) (49.717) (50.027) (49.78) (47.845)
NTC 72.855 70.856 68.330 69.980 65.980 66.836 65.102 64.792 70.160 77.126
(72.747) (71.164) (68.691) (69.365) (68.097) (68.597) (69.784) (69.137) (69.303) (68.361)
CCC 97.723 94.569 90.645 92.387 86.562 86.710 83.998 85.970 88.427 94.165
(81.625) (80.204) (76.9) (77.745) (75.093) (75.544) (76.933) (76.968) (75.979) (75.875)
CATA 0.499 0.508 0.519 0.527 0.528 0.530 0.522 0.530 0.547 0.569
(0.188) (0.191) (0.193) (0.194) (0.194) (0.195) (0.196) (0.195) (0.194) (0.193)
Downloaded by Reserve Bank of India At 23:18 27 August 2017 (PT)

CLTA 0.418 0.426 0.430 0.419 0.413 0.415 0.406 0.401 0.415 0.409
(0.183) (0.182) (0.184) (0.184) (0.18) (0.181) (0.179) (0.181) (0.184) (0.184)
SIZE 8.864 8.926 9.024 9.095 9.184 9.252 9.256 9.282 9.330 9.334
(1.422) (1.436) (1.43) (1.439) (1.422) (1.427) (1.436) (1.434) (1.44) (1.46)
LR 0.155 0.146 0.145 0.147 0.148 0.141 0.142 0.133 0.124 0.126
(0.142) (0.14) (0.138) (0.141) (0.142) (0.139) (0.14) (0.138) (0.131) (0.134)
Table I. CACL 1.364 1.354 1.373 1.431 1.445 1.439 1.444 1.496 1.482 1.564
Year wise average (0.734) (0.721) (0.721) (0.743) (0.742) (0.726) (0.74) (0.757) (0.738) (0.774)
of variables Note: Figures in parentheses denote SD

due to hampering of business activity post financial crisis. Turning to working capital
measures, it is observed that ITD declined from 97 days in 2003 to 86 days in 2007. However,
post 2007 its movement has been erratic with no clear trend. Likewise, ACP and APP levels
have also registered consistent drop during 2003 to 2007. CCC reduced from 97 days in 2003 to
86 days in 2008 displaying reduction in receivable period. However, the figure again increased
to 94 days in 2012. The general behavior of CCC is in tandem with other indicators of working
capital management. The logarithm of average sales size of firms studied has grown
continuously from 8.8 in 2003 to 9.2 in 2008 and further to 9.3 in 2012. Financial indicators
have shown mixed results, with average LR declining to 12 percent in 2012 from 15 percent in
2003 indicating fall in debt financing. The average liquidity of Indian firms suggests
improvement with CACL roughly increasing from 1.3 in 2003 to 1.5 in 2012.
The firms covered in existing study encompass many sectors. Table II imparts major
industry-wise summary of variables. It may be noted that largest number of firms belongs
to chemical products followed by textiles. Construction activity comprises least number of
firms. Focusing on profitability indicators, it is noted that computer related firms are most
profitable with GOPR at 68 percent implying a broad margin. Machinery and accounting
comes next in profitability with average GOPR for period at 52 percent. The lowest
profitability is recorded by textile sector at 21 percent with low standard deviation, implying
high competition and narrow margin in textile sector. Turning to working capital
management it emerges that firms belonging to construction undergo longest period to
convert inputs to outputs with most indicators, namely, ITD, APP, NTC, CCC being
unanimously highest across industries at 117, 97, 118 and 143 days, respectively. It is
followed by Machinery and accounting with average CCC of 118 days. On the other hand,
motor vehicles and transport have displayed shortest lag of conversion period of 39 and
54 days as per NTC and CCC, respectively. Sales size-wise, iron and steel firms have been
noted to be of largest size followed by firms belonging to motor vehicles and transport.
Financial soundness-wise, textile sector emanates as most leveraged group registering an
Downloaded by Reserve Bank of India At 23:18 27 August 2017 (PT)

Food Iron Machinery and Motor Trade


Chemical Electrical products and and accounting vehicles and wholesale
products Computer Construction machinery beverages steel machine transport Textiles and retail Manufacturing Services

Number
of firms 2,052 417 248 575 324 441 932 637 941 378 9,693 1,555
GOPR 0.435 0.686 0.364 0.442 0.372 0.287 0.519 0.444 0.208 0.425 0.384 0.498
(0.281) (0.362) (0.293) (0.279) (0.255) (0.213) (0.287) (0.239) (0.147) (0.322) (0.267) (0.34)
ITD 91.935 34.635 117.105 95.530 75.834 88.248 105.374 67.576 90.313 70.145 94.407 50.260
(51.059) (53.849) (80.145) (51.806) (50.352) (44.798) (56.296) (46.213) (53.619) (59.425) (54.806) (59.75)
ACP 75.358 112.050 101.762 102.328 53.637 50.321 92.376 60.405 50.783 75.991 70.249 79.522
(43.736) (50.372) (52.918) (49.102) (45.232) (35.056) (45.833) (36.846) (39.586) (51.765) (46.147) (55.356)
APP 75.377 69.357 97.104 81.511 65.594 77.760 84.320 74.467 47.671 78.000 71.965 85.695
(46.017) (55.186) (58.626) (50.776) (56.351) (49.551) (44.87) (36.074) (41.542) (60.061) (47.978) (61.825)
NTC 69.639 67.864 118.391 98.288 54.667 45.098 92.482 39.923 80.005 58.328 73.990 37.524
(61.277) (78.378) (77.569) (64.607) (60.688) (56.894) (64.089) (53.802) (57.583) (72.754) (65.587) (79.25)
CCC 93.068 78.637 143.846 118.065 72.279 61.169 118.380 54.969 96.592 72.287 95.588 53.180
(68.861) (87.389) (90.484) (71.87) (69.277) (56.922) (70.893) (62.998) (66.16) (76.615) (73.275) (85.692)
CATA 0.523 0.524 0.738 0.611 0.496 0.489 0.641 0.496 0.428 0.686 0.527 0.510
(0.182) (0.194) (0.132) (0.174) (0.19) (0.173) (0.167) (0.153) (0.16) (0.166) (0.186) (0.224)
CLTA 0.402 0.285 0.555 0.415 0.406 0.472 0.451 0.456 0.402 0.436 0.427 0.336
(0.166) (0.178) (0.158) (0.177) (0.188) (0.193) (0.182) (0.145) (0.171) (0.204) (0.177) (0.192)
SIZE 9.298 8.629 9.674 9.115 8.673 9.971 8.979 9.830 9.141 8.616 9.249 8.568
(1.398) (1.586) (1.357) (1.453) (1.417) (1.026) (1.554) (1.023) (1.288) (1.585) (1.387) (1.574)
LR 0.124 0.038 0.117 0.113 0.137 0.191 0.068 0.147 0.246 0.066 0.148 0.100
(0.127) (0.074) (0.101) (0.113) (0.131) (0.142) (0.097) (0.109) (0.15) (0.096) (0.14) (0.131)
CACL 1.446 2.093 1.464 1.680 1.436 1.148 1.592 1.195 1.182 1.802 1.386 1.735
(0.709) (0.91) (0.561) (0.73) (0.782) (0.574) (0.698) (0.577) (0.627) (0.769) (0.701) (0.907)
Note: Figures in parentheses denote SD
Working

577
capital
management

Major industry-wise
Table II.

summary
JES average LR of 24 percent followed by iron and steel at 19 percent. Computer related firms
44,4 emerged as least leveraged at 3.8 percent but highly liquid at 209 percent. Iron and steel
turns to be least liquid with paltry 114 percent.
Correlation coefficient has been calculated to examine linear association amongst
variables of interest (Table III). CCC is positively associated with ITD, ACP and NTC,
implying movement in tandem with all working capital indicators[3]. Greater conversion
578 period is having strong detrimental association with profitability indicator, namely, GOPR.
Table IV displays the regression results estimating Equation (2). Balanced panel has
been employed comprising of 1,172 firms for ten years leading to 11,720 total numbers of
observations. GOPR constitutes our dependent variable. Model 1 constitutes the standard
fixed effects regression. Hausman test was performed that provided results in favor of FEM.
The impact of working capital management as measured by CCC is significantly negative
on corporate profitability. The finding is in conjunction with Deloof (2003) for Belgian firms
or Pakistan firms as carried out by Raheman et al. (2010). The coefficient of SIZE is positive
Downloaded by Reserve Bank of India At 23:18 27 August 2017 (PT)

and significant indicating larger firms to be more profitable. Amongst financial ratios, both
current ratio (CACL) and CATA have significant positive impact on GOPR. The result
signifies that as company’s ability to pay back its short-term liabilities improves so does its
performance. Conversely, both LR and CLTA are strongly inversely related to dependent
variable implying retarding influence on firm profits.

GOPR ITD ACP APP NTC CCC CATA CLTA SIZE LR CACL

GOPR 1.000
ITD −0.279 1.000
ACP −0.002 0.164 1.000
APP −0.034 0.202 0.320 1.000
NTD −0.264 0.562 0.523 −0.286 1.000
CCC −0.212 0.692 0.509 −0.194 0.958 1.000
CATA 0.256 0.162 0.344 0.053 0.333 0.302 1.000
CLTA −0.086 0.071 0.130 0.245 0.029 −0.002 0.317 1.000
Table III. SIZE 0.183 −0.086 −0.225 −0.105 −0.155 −0.152 −0.052 0.006 1.000
Correlation coefficient LR −0.344 0.054 −0.124 0.019 −0.044 −0.030 −0.404 −0.093 0.192 1.000
amongst variables CACL 0.233 0.039 0.150 −0.161 0.207 0.211 0.407 −0.615 −0.110 −0.222 1.000

Fixed effects Robust Bayesian REM Bayesian FEM


SE (Model 1) (Model 2) (Model 3)

Dependent variable: gross operating profit ratio (GOPR)


Intercept 0.191 (0.066)* 0.302 (0.087)* 0.418 (0.036)*
CCC −0.181 (0.023)* −0.178 (0.013)* −0.185 (0.014)*
Log of sales (SIZE) 0.0198 (0.007)* 0.019 (0.004)* 0.043 (0.065)*
Debt to total assets ratio (LR) −0.251 (0.028)* −0.251 (0.019)* −0.221 (0.054)*
Current assets to current liabilities ratio (CACL) 0.044 (0.008)* 0.045 (0.006)* 0.05 (0.03)*
Current liabilities to total assets ratio (CLTA) −0.076 (0.035)** −0.076 (0.025)* −0.035 (0.153)*
Current assets to total assets (CATA) 0.456 (0.034)* 0.455 (0.02)* 0.453 (0.027)*
Model statistics
Adj. R2 0.2576 0.6532 0.7056
Table IV. σ 0.195 300.6* 0.130*
Estimation results Notes: Figures in parenthesis denote SE. *,**Significant at 1 and 5 percent levels, respectively
Model 2 and Model 3 show estimates of Equation (2) applying Bayesian technique assuming Working
random and fixed effects, respectively, with normal likelihood distribution. It is observed capital
that model selection criterion adjusted R2 strongly supports fixed effects Bayesian management
compared to random effects. At a broader level Model 1 results are similar to Bayesian
regressions. The marginal impact of SIZE is more profound in Model 3 as compared to
Model 1 and Model 2. To have a visual glance of Monte Carlo draws, probability density and
autocorrelation plot of parameters for Model 3 are provided in Figure A1. Plots indicate that 579
all parameters of the chain are mixing well with autocorrelation vanishing before 50 lags in
each case. Posterior density of parameters is almost converging to normal distribution.
In linear regressions, adjusted R2 is calculated to quantify variation in response variable
that is explained by explanatory variables. Although such statistics in not directly available
in mixed models, alternative R2 has been suggested in literature that are widely used in
case of mixed models and help to assess comparability and suitability of mixed models
(Xu, 2003; Gelman and Pardoe, 2006). The adjusted R2 has shown considerable improvement
Downloaded by Reserve Bank of India At 23:18 27 August 2017 (PT)

from 0.25 in Model 1 to 0.65 in Model 2 and further to 0.70 in Model 3 demonstrating
significant gains owing to Bayesian estimation.
As elaborated in Section 4, more general multivariate student t-distribution, i.e. a small
sample distribution has also been applied and its output tabulated in Table V. The student
t-distribution enables us to capture skewness more appropriately in data. Varied measures
of working capital have been employed to assess their impact on profitability and examine
robustness of results. It is observed that working capital measures, NTC, ITD, ACP and

Model 4 Model 5 Model 6 Model 7 Model 8 Model 9

Dependent variable: gross operating profit ratio (GOPR)


Intercept 0.425 0.423 0.419 0.424 0.378 0.395
(0.039)* (0.046)* (0.028)* (0.089)* (0.067)* (0.052)*
CCC −0.160 0.1088
(0.015)* (0.070)
CCC2 0.0806
(0.143)
APP 0.042
(0.021)**
NTC −0.316
(0.015)*
ITD −0.300
(0.026)*
ACP −0.075
(0.030)**
Log of sales (SIZE) 0.028 0.021 0.021 0.029 0.023 0.02589
(0.010)* (0.009)** (0.012)*** (0.011)* (0.010)* (0.031)
Debt to total assets ratio (LR) −0.241 −0.24 −0.225 −0.239 −0.238 −0.233
(0.019)* (0.019)* (0.020)* (0.020)* (0.019)** (0.029)*
Current assets to current 0.044 0.038 0.038 0.045 0.041 0.043
liabilities ratio (CACL) (0.007)* (0.007)* (0.008)* (0.008)* (0.007)* (0.014)*
Current liabilities to total assets −0.07 −0.098 −0.058 −0.063 −0.083 −0.071
ratio (CLTA) (0.030)** (0.030)* (0.035)*** (0.033)*** (0.032)* (0.067)
Current assets to total assets 0.443 0.48 0.431 0.437 0.458 0.456
(CATA) (0.020)* (0.020)* (0.020)* (0.021)* (0.020)* (0.022)*
Model statistic Table V.
σ 0.131* 0.127* 0.131* 0.131* 0.128* 0.130* Bayesian estimation
Adj R2 0.794 0.802 0.797 0.794 0.796 0.789 results with student
Notes: Figures in parenthesis denote SE. *,**,***Significant at 1, 5, and 10 percent levels, respectively t-distribution
JES CCC are having strong negative impact on gross operating profit ratio. It signifies lower
44,4 profitability with increase in trading cycle. On contrary, APP has positive impact on GOPR.
More a firm is able to delay its payments, more funds it has at its disposal for other
productive activities that improves profitability.
Equation (3) has also been estimated that explores if non-linearity between CCC and GOPR
has a role in the model, tabulated under Model 9. It is found that inclusion of CCC2 leads to loss
580 of significance for both CCC and CCC2 due to which nothing can be decisively concluded for
these working capital indicators. Amongst financial indicators, LR has a robust inverse
relation across all models. In accordance with regression results with normal errors (Table IV),
current assets ratio (CATA) has beneficial influence on GOPR. Overall, results portray robust
inverse relationship between working capital management and profitability for study period
both across varied measures of working capital management and regression methodologies.
Major industry-wise Bayesian regression has been performed and displayed in Table VI. It is
observed that both the impact of SIZE and LR is insignificant for firms belonging to wholesale
Downloaded by Reserve Bank of India At 23:18 27 August 2017 (PT)

and retail trade implying both size and external debt to be weak determinants of profitability.
The liquidity ratio CACL is insignificant for Model 1 and Model 2 indicating effect of liquidity to
be weak for both sub-industries chemicals and machinery. Overall results obtained are
qualitatively similar as obtained based on Table V pointing robust estimates.

6. Conclusion
Working capital management is pivotal for short-term financial decisions. Suitable working
capital is desired for optimal firm performance. There exist numerous studies examining
impact of long term capital structure and investment on firm profitability. However, studies
focusing on relationship between company short-term liquidity needs on performance are
limited for Indian context. Bridging this gap, existing analysis has endeavored to examine
impact of working capital on profitability for Indian corporate entities. A rich firm level
balanced panel spanning from 2003 to 2012 is constructed to examine propositions for
Indian corporate sector encompassing varied industries like chemicals, iron and steel,
machinery, computers to name a few. Varied measures of working capital are constructed
like CCC, inventory turnover in days, average collection period, net trading cycle.
Correlation analysis among working capital indicators displays strong procyclicality.
The study applies standard panel data regression and Bayesian methodology with priors
distributed as more general multivariate student t-distribution for robust statistical inference.

Model 1: chemical Model 2: machinery and Model 3: wholesale and


Dependent variable: GOPR products accounting retail trade

Intercept −3.344 (0.008)* −2.157 (0.004)* 4.077 (0.014)*


CCC −0.252 (0.030)* −0.538 (0.058)* −0.233 (0.052)*
Log of sales (SIZE) 0.061 (0.010)* 0.1190 (0.017)* 0.013 (0.014)
Debt to total assets ratio (LR) −0.305 (0.037)* −0.233 (0.089)* 0.061 (0.093)
Current assets to current liabilities
ratio (CACL) 0.009 (0.012) 0.021 (0.028) 0.047 (0.027)***
Current liabilities to total assets
ratio (CLTA) −0.254 (0.049)* −0.156 (0.016)* −0.058 (0.017)*
Current assets to total assets
(CATA) 0.304 (0.043)* 0.420 (0.095)* 0.046 (0.091)
Table VI.
Industry wise results- Model statistic
Bayesian FEM, Adj. R2 0.647 0.653 0.705
t distribution Notes: Figures in parenthesis denote SE. *,***Significant at 1, and 10 percent levels, respectively
To ascertain model fit and comparison, adjusted R2 statistics has been derived for Working
Bayesian regressions also. The exercise reveals considerable gains in overall model fit as capital
obtained due to application of multivariate student t-priors. It is revealed that stochastic part management
of model is more appropriately captured by exact distribution using Bayesian approach.
Analytical results depict that longer cash conversion period has detrimental influence on
profitability for Indian firms during study period across Bayesian regression methodologies.
A suitable level of working capital is strongly evidenced across all working capital measures. 581
Financial soundness indicators are playing significant role in determining firm profitability.
Larger firms are more profitable whereas higher leverage has negative influence on
firm performance.

Notes
1. CCC stands for Cash Conversion Cycle that is standard measure for working capital. It quantifies
Downloaded by Reserve Bank of India At 23:18 27 August 2017 (PT)

length of time a concern takes to convert inputs to cash flows. Further formula details provided in
Section 3.
2. See Eljelly (2004) and Lazaridis and Tryfonidis (2006).
3. All correlation coefficient figures are found to be significant at 1 percent level.

References
Amisano, G. and Giorgetti, M.L. (2013), “Entry into pharmaceutical submarkets: a Bayesian panel
probit analysis”, Journal of Applied Econometrics, Vol. 28 No. 4, pp. 667-701.
Bauer, R., Cosemans, M., Frehen, R. and Schotman, P. (2008), “A Bayesian panel data approach to
explaining market beta dynamics”, Discussion Paper No. 03/2008-051, Tilburg University.
Deloof, M. (2003), “Does working capital management affect profitability of Belgian firms?”, Journal of
Business Finance and Accounting, Vol. 30 Nos 3-4, pp. 573-587.
Eljelly, A. (2004), “Liquidity-profitability tradeoff: an empirical investigation in an emerging market”,
International Journal of Commerce and Management, Vol. 14 No. 2, pp. 48-61.
Filbeck, G. and Krueger, T.M. (2005), “An analysis of working capital management results across
industries”, American Journal of Business, Vol. 20 No. 2, pp. 11-18.
Gelman, A. and Pardoe, I. (2006), “Bayesian measures of explained variance and pooling in multilevel
(hierarchical) models”, Technometrics, Vol. 48 No. 2, pp. 241-251.
Ghosh, S.K. and Maji, S.G. (2004), “Working capital management efficiency: a study on the Indian
cement industry”, The Management Accountant, Vol. 39 No. 5, pp. 363-372.
Harris, A. (2005), “Working capital management: difficult, but rewarding”, Financial Executive, Vol. 21
No. 4, pp. 52-53.
Koop, G., Poirier, D.J. and Tobias, J.L. (2007), Bayesian Econometric Methods, Vol. 7, Cambridge
University Press, New York, NY.
Kumar, P. and Shrivastava, A. (2013), “Trade credits and its impact on corporate performance: some
empirical findings in Indian context”, Journal of Management Value and Ethics, Vol. 3 No. 3,
pp. 41-57.
Lange, K.L., Little, R.J. and Taylor, J.M. (1989), “Robust statistical modeling using the t distribution”,
Journal of the American Statistical Association, Vol. 84 No. 408, pp. 881-896.
Lazaridis, I. and Tryfonidis, D. (2006), “Relationship between working capital management and
profitability of listed companies in the Athens stock exchange”, Journal of Financial
Management and Analysis, Vol. 19 No. 1, pp. 26-35.
Levine, R. (2005), “Finance and growth: theory and evidence”, Handbook of Economic Growth, Vol. 1,
Part A, pp. 865-934.
JES Levine, R., Loayza, N. and Beck, T. (2000), “Financial intermediation and growth: causality and causes”,
44,4 Journal of monetary Economics, Vol. 46 No. 1, pp. 31-77.
Meligkotsidou, L., Tzavalis, E. and Vrontos, I.D. (2012), “A Bayesian panel data framework for
examining the economic growth convergence hypothesis: do the G7 countries converge?”,
Journal of Applied Statistics, Vol. 39 No. 9, pp. 1975-1990.
Mitchell, J., Pain, N. and Riley, R. (2011), “The drivers of international migration to the UK: a panel-
582 based Bayesian model averaging approach”, The Economic Journal, Vol. 121 No. 557,
pp. 1398-1444.
Moral-Benito, E. (2012), “Determinants of economic growth: a Bayesian panel data approach”, Review of
Economics and Statistics, Vol. 94 No. 2, pp. 566-579.
Raheman, A., Afza, T., Qayyum, A. and Bodla, M.A. (2010), “Working capital management and
corporate performance of manufacturing sector in Pakistan”, International Research Journal of
Finance and Economics, Vol. 47 No. 1, pp. 156-169.
Shin, H. and Soenen, L. (1998), “Efficiency of working capital and corporate profitability”, Financial
Downloaded by Reserve Bank of India At 23:18 27 August 2017 (PT)

Practice and Education, Vol. 8 No. 2, pp. 37-45.


Xu, R. (2003), “Measuring explained variation in linear mixed effects models”, Statistics in Medicine,
Vol. 22 No. 22, pp. 3527-3541.
Zellner, A. and Ando, T. (2010), “Bayesian and non-Bayesian analysis of the seemingly unrelated
regression model with student-t errors, and its application for forecasting”, International Journal
of Forecasting, Vol. 26 No. 2, pp. 413-434.
Appendix 1 Working
capital
management

 1
Autocorrelation

Autocorrelation
1.0 1.0

0.0 0.0
583
–1.0 –1.0
0 50 0 50
lag lag

3 4
Autocorrelation

Autocorrelation
1.0 1.0

0.0 0.0

–1.0 –1.0
Downloaded by Reserve Bank of India At 23:18 27 August 2017 (PT)

0 50 0 50
lag lag

5 6
Autocorrelation

Autocorrelation

1.0 1.0

0.0 0.0

–1.0 –1.0
0 50 0 50
lag lag

7 
Autocorrelation

Autocorrelation

1.0 1.0

0.0 0.0

–1.0 –1.0
0 50 0 50
lag lag

 sample: 8,000 1 sample: 8,000


40.0
P (1)
P ( )

20.0
20.0

0.0 0.0
0.35 0.4 0.45 0.5 –0.25 –0.2 –0.15
 1

3 sample: 8,000 4 sample: 8,000


80.0
P ( 3)

P ( 4)

20.0
40.0

0.0 0.0
–0.35 –0.3 –0.25 –0.2 0.02 0.04 0.06
3 4

5 sample: 8,000 6 sample: 8,000


20.0 20.0
P ( 5)

P ( 6)

10.0 10.0

0.0 0.0
–0.2 –0.15 –0.1 –0.05 0.35 0.4 0.45 0.5
5 6

7 sample: 8,000  sample: 8,000


0.8
P ( 7)

P ()

100.0
0.4

0.0 0.0
0.09 0.1 0.11 40.0 42.0 44.0
Figure A1.
7  Autocorrelations plots
and Kernel density
Note: 1, 2, 3, 4, 5, 6, and 7 correspond to CCC, SIZE, of the parameters
LR, CACL, CLTA, CATA, GWCTR coefficients, respectively
JES Appendix 2. Calculation of posterior distribution of regression coefficients and
44,4 individual effects
Let us consider the panel regression model given as follows:
 
yit ¼ X it b þai þ Eit where Ei  N 0; t1 (A1)
On stacking over time fixed-effect model can be re-written as follows:
 
yi ¼ X i b þai þ Ei where Ei  t 0; t1 I T (A2)
584
 
y ¼ X bþ aþ E; where E  t 0; t1 I N T (A3)
The likelihood function of the fixed-effect model under the assumption of normal distribution is given
as follows:
  n t o
ðyX baÞ0 ðyX baÞ
NT
L y=K; t pt 2 exp  (A4)
2
The prior distributions of unknown parameters β, a and τ are given by:
Downloaded by Reserve Bank of India At 23:18 27 August 2017 (PT)

 t k=2   h    i
pðbÞ ¼ Lb 1=2 exp t bb0 0 Lb bb0 (A5)
2p 2
 t k=2 h t i
pð a Þ ¼ jLa j1=2 exp  ðaa0 Þ0 La ðaa0 Þ (A6)
2p 2
t
ta0 1 e b0
pðtÞp ; 0 oto1 (A7)
Gða0 Þb0 a0
where β0, α0, a0 and b0 are the fixed constants.
Combining the likelihood function with prior probabilities and assuming the errors distributions
follows multivariate normal density, the posterior density of β is given by:
ZZ  N T  t 
   t 2
p by p exp  ðyX baÞ0 ðyX baÞ
2p 2
 t k21  1       k2  
Lb 2 exp t bb0 0 Lb bb0  t 2 jLa j12 exp ðaa0 Þ0 La ðaa0 Þ ta0 1 eb0 dtda
t

2p 2 2p
(A8)
Similarly, combining the likelihood function with prior probabilities and assuming the errors
distribution follows multivariate t density, the posterior density of β is given by:
ZZ
    1 h t iu þ2N T
p by p t1  2 1 þ ðyX bW aÞ0 ðyX bW aÞ
u
 t k21  1       k2  
Lb 2 exp t bb0 0 Lb bb0  t 2 jLa j12 exp ðaa0 Þ0 La ðaaÞ ta0 1 eb0 dtda (A9)
t

2p 2 2p
Integrating over τ and α leads to the marginal posterior distribution of β. In the similar fashion, we can
get the posterior distribution of the individual effect α on integration over β and τ of the Equations (A8)
and (A9).

Corresponding author
Nitin Kumar can be contacted at: nitin_005us@yahoo.com

For instructions on how to order reprints of this article, please visit our website:
www.emeraldgrouppublishing.com/licensing/reprints.htm
Or contact us for further details: permissions@emeraldinsight.com

View publication stats

You might also like