Sie sind auf Seite 1von 8

International Research Journal of Finance and Economics

ISSN 1450-2887 Issue 46 (2010)


© EuroJournals Publishing, Inc. 2010
http://www.eurojournals.com/finance.htm

The Evolutionary Trends of Performance Measures:


An Application of Power Law

Chia-Sheng Hung
Department of Accounting and Information Science
Nan Hua University, ChiayiCounty, Taiwan, R.O.C
E-mail: eco0303@gmail.com

Yu-Chung Hung
Department of Accounting and Information Technology
National Chung Cheng university, ChiayiCounty
Taiwan, R.O.C
E-mail: actych@ccu.edu.tw

Abstract

A good performance measure should discriminate which programs, methods, or


employees are effective and efficient. This paper first takes advantage of Power law to
examine the distributions of measures and point out which measures are more
discriminating. Comparing with other financial measures, our empirical results indicate that
ROA is more discriminating. In addition, ROA is more discriminating over time. Hence,
ROA is a good performance measures.

Keywords: Power law, Zipf lwa, performance measure, financial performance


JEL Classification Code: M10

1. Introduction
Resources are limited. As a result, people strive to conserve the economic resources and make sure that
they are used effectively. In order to reach the desired results, organizations evaluate performance and
establish parameters within programs, investments, and acquisitions. Although the numbers of
measures are not directly connected to improving performance, they represent the means to achieve the
ultimate purpose. Hence, performance measures play the important role to provide information and
guide the management/investors to draw strategies and arrange the portfolio.
Within an organization, a good performance measure can help organization to evaluate
performance, control budget, motivate employees and so on. A good performance measure can not
only provide a common language for communication, but also provide a way to see if the strategy is
working1. If a measure is sensitive with time, it can not provide the true information for the top
management. The main purpose of this paper is to investigate the sensitivity with time of performance
measures by take advantage of Power law.
The Power law is widely recognized as a universal law in describing many physical phenomena
and drafted by Pareto (1897). Pareto (1897) first report on the individual income distribution and he

1
Enterprise Architecture Program (2007). Treasury IT Performance Measures Guide. U.S. Department of the Treasury.
May 2007.
131 International Research Journal of Finance and Economics - Issue 46 (2010)

showed that the probability density distribution of income follows a power law distribution in the high-
income range. The range of individual income is reported to be approximated by a log normal
distribution or an exponential distribution (Montroll and Shlesinger, 1983; Aoyama et al., 2000;
Dragulescu and Yakovenko, 2001) Zipf (1949) extend the use of Power law beyond the physical
sciences and showed the law was applicable to the social science. Prior studies have shown the
financial variables such as firm size, stock returns, and net income which were fit for the Power law
(Axtell, 2001; Gabaix et al., 2003; Okuyama et al., 1999). By estimating the parameter of Power law,
we can examine the trends of performance measures and find which measure is more sensitive with
time.
Section 2 provides a review of the literature on Power law and its extension. Section 3 presents
the research model and methodology. The sample and empirical results are discussed in Section 4.
Section 5 presents the sensitivity analysis and conclusions respectively.

2. Literature review
The Power law is widely recognized as a universal law in describing many physical phenomena and
drafted by Pareto (1897). Pareto (1897) first report on the individual income distribution and he
showed that the probability density distribution of income follows a Power law distribution in the high-
income range. Zipf (1949) extended the use of the Power law beyond the physical sciences and showed
that the law was applicable to the social sciences. The Power law can be applied to any item that can be
ranked by size. The Power law is stated as:
( Size) i * ( Rank ) iq = Cons tan t
The exponent “q” is known as the Power law exponent or Zipf’s parameter. Zipf law is a
special case of Power law when the exponent “q” is equal to 1. There are several literatures shows that
the distribution of financial variables follows Power law. For examples, Axtell (2001) demonstrate the
firm size follows Power law. Gabaix et al. (2003) and Okuyama et al. (1999) report respectively the
stock returns and net income follow Power law. Some similar results are reported in Ramsden and
Kiss-Haypal (2000) and Stanley et al. (1995) in terms of assets and sales. Some literatures employ the
Power law to model the growth of firms, they show the distribution of the logarithm of growth rates
has an exponential form and the fluctuations in the growth rates scale with the firm size (Stanley et al.,
1996; Buldyrev et al., 1997; and Mizuno et al., 2004). There are some applications of Power law
presented in other fields of researches. Bosker, et al. (2008), Chen and Zhou (2004) and Lucien and
Blumenfeld-Lieberthal (2007) employ the Power law to model the city size and Mansilla et al. (2007)
apply Power law to the behavior of journal impact factors.
Naldi (2003) derive the relationship between concentration indices and Zipf’s parameter to
describe the unevenness in economics. Naldi evaluated the sensitivity of the concentration indices,
including Gini, Bonferroni and Amato, and the Hirschman-Herfindahl Index (HHI), to changes in
Zipf’s parameter to assess the resolution capabilities of the concentration indices. From his results, the
Hirschman-Herfindahl Index (HHI) is the most sensitive index in contexts where Zipf’s law applies.
Balakrishnan et al. (2008) model the distribution of daily trading volume across all stocks in the U.S.
market and in each of the three U. S. exchanges as a Power law function. They found the Power law
exponent systematically increases with time and the exponents evolve for the market and for each of
the exchanges over the years 1962 to 2005. Their finding indicates that the trading is becoming
increasingly concentrated in a subset of stocks rather than in all stocks. Alegria and Schaeck (2008)
investigate the effect of changes in Zipf’s exponent and the bank sizes on the behavior of different
concentration indices, such as 3-bank concentration ratio, the Hirschman-Herfindahl Index (HHI) and
the top 5%-concentration ratio. Alegria and Schaeck’s works investigate the elasticity of the three
concentration indices to changes in sample size and Zipf’s exponent and they found the top 5%-
concentration ratio is least sensitive to changes in sample size and Zipf’s exponent. Hence, the top 5%-
concentration ratio is the most suitable index for estimating the concentration of banking systems.
International Research Journal of Finance and Economics - Issue 46 (2010) 132

Zipf (1949) is the first work that viewed language as a “tool” that is shaped by its “jobs” in
human society. Zipf proposed that the behaviors which are “useful” are performed frequently, and
frequent behaviors become quicker and easier to perform. Basing on the Zipf (1949)’s finding,
examined the relationship between rank and frequency of various linguistic and social units and
constructions. Huang et al. (2008) develop an innovative fraud detection mechanism on the basis of
Zipf’s law to assist auditors in reviewing the overwhelming volumes of datasets and identifying any
potential fraud records. The results demonstrate that Zipf analysis can assist auditors to locate the
source of suspicion and future enhances the resulting audit processes.

3. Data and Performance Measures


Basically, the performance indicators include accounting performance indicators and market
performance indicators (Lee, 2006). The accounting performance refers to the profitability and cost
measures. This paper focuses on the profitability measures. By following Edvinsson and Malone
(1997), we employ return to assets (denoted as ROA), the ratio of operating income to sales (denoted
as OITS), and the ratio of net assets value to assets (denoted as NAVT) as performance measures. This
paper examines the financial measures on banking industry in Taiwan. The data used in this paper are
mainly from the Bureau of Monetary Affairs, Financial Supervisory Commission in Taiwan. These
data include 35 banks. In order to avoid the effects of financial tsunami on the empirical results, the
empirical period is from 1992 to 2005.

4. The Evolutionary Trends of Financial Measures


In this section, we model the distribution of the performance measure as a Power law function and
estimate the value of the Power law exponent. In addition, we investigate if the Power law exponent
fluctuates over time, and if so, the implications of such change on the enterprises’ booms or slumps are
not in concert at the same time.
The Power law is stated as:
( Size) it * ( Rank ) itqt = Cons tan t t (Model 1)
The exponent “q” is known as the Power law exponent or Zipf’s parameter. Zipf law is a
special case of Power law when the exponent “q” is equal to 1. If the q increases of a performance
measure, it represents that the distribution of data becomes more concentrated. Hence, it will be more
difficult to discriminate the performance of agents. The subscriptions of i and t denote the firm i and
period t . Equation (1) can be stated as log-form:
log(Size) it + q log( Rank ) it = log(Cons tan t t ) (Model 2)
So we can obtain the following equation:
log(Size) it = cons tan t t − qt log( Rank ) it (Model 3)
By using OLS, we can estimate the value of qt for every single period. Hence, we will have t
values of the exponent “ q ”. The estimated exponents “ q ” of performance measures are shown in
Table 1. We can find that the Power law exponents of ROA fluctuate from -1.288 to -0.151. Only the
exponent of year 2005 is not significantly different from zero. The Power law exponents of OITS are
between -4.595 and -0.981. The exponents of year 2000 to 2005, except year 2004, are not
significantly different from zero. The Power law exponents of NAVT fluctuate from -1.082 to -0.666.
Every Power exponents of NAVT is significantly different from zero. The exponents of year 1992 to
2003 are close to 1. This phenomenon indicates that Zipf law stands in this period for NAVT. The
average exponents of ROA, OITS, and NAVT are -0.515, -0.782, and -0.943. Hence, the distribution of
NAVT is more concentrated than the other two measures’ distributions. Meanwhile, it is easier to use
ROA to discriminate bank’s performance.
133 International Research Journal of Finance and Economics - Issue 46 (2010)

Table 1: The Power law exponents ( q ) of performance measures

ROA OITS NAVT


year R_Square q R_Square R_Square
-1.288*** -0.981*** -1.082***
1992 0.780 0.647 0.723
(-11.039) (-8.412) (-9.999)
-1.066*** -0.672*** -1.071***
1993 0.899 0.923 0.853
(-16.876) (-21.015) (-15.065)
-0.814*** -0.654*** -1.009***
1994 0.857 0.609 0.916
(-15.100) (-7.755) (-20.685)
-0.128*** -0.329*** -0.986***
1995 0.769 0.101 0.936
(-11.006) (-2.275) (-23.912)
-1.117*** -0.855*** -0.965***
1996 0.897 0.378 0.924
(-16.871) (-4.973) (-21.838)
-0.655*** -0.599*** -0.882***
1997 0.796 0.373 0.888
(-12.224) (-4.860) (-17.648)
-0.378*** -0.474*** -0.926***
1998 0.424 0.155 0.858
(-5.455) (-2.888) (-15.604)
-0.351*** -0.349*** -0.917***
1999 0.463 0.110 0.864
(-5.962) (-2.443) (-15.990)
-0.189** -0.291 -0.934***
2000 0.094 0.012 0.867
(-2.017) (-1.221) (-16.183)
-0.277** -0.264 -0.958***
2001 0.072 0.026 0.876
(-2.001) (-1.431) (-16.832)
-0.273*** -0.313 -1.080***
2002 0.099 0.022 0.849
(-2.295) (-1.367) (-14.834)
-0.340*** -0.292* -1.019***
2003 0.293 0.055 0.910
(-4.147) (-1.808) (-19.826)
-0.151*** -4.595*** -0.700***
2004 0.630 0.362 0.419
(-7.995) (-4.693) (-5.256)
-0.180 -0.282 -0.666***
2005 0.045 0.020 0.925
(-1.643) (-1.311) (-21.089)
Average -0.515 -0.782 -0.943
PS: the values in parentheses are t-value; *, ** and *** represent the significant level 0.1, 0.05 and 0.01, respectively.

Table 2 shows the Power law exponents of the growth rates of performance measures. The
results are similar to the table 1. The average exponents of growth rate of ROA, growth rate of OITS,
and growth rate of NAVT are -0.333, -0.916, and -0.804. The average exponents of Hence, the
distribution of growth rate of OITS is more concentrated than the other two growth rates’ distributions.
The absolute value of average exponents of growth rate of ROA is smaller than two other growth rates’
absolute values. It is easier to use growth rate of ROA to discriminate bank’s growth.
Figure 1 and figure 2 demonstrate the trends of various exponents. We can see that the
exponents of OITS and its growth rate are outliers in year 2004. This phenomenon may be caused by
the second financial reform spurred by Taiwan government. Several banks merged and some banks
suffered serious loss and failed in 2004. Hence, the operating income fluctuated furiously.
International Research Journal of Finance and Economics - Issue 46 (2010) 134
Table 2: The Power law exponents ( q ) of performance growth rate

Growth rate Growth rate Growth rate


of ROA of OITS of NAVT
year R_Square q R_Square R_Square
-0.392*** -0.956*** -1.165***
1992 0.507 0.457 0.836
(-5.907) (-5.744) (-13.962)
-0.626*** -1.384*** -0.958***
1993 0.802 0.647 0.490
(-11.444) (-8.178) (-6.122)
-0.472*** -0.410*** -1.084***
1994 0.653 0.131 0.849
(-7.829) (-2.532) (-14.869)
-0.156 -0.302 -0.508***
1995 0.026 0.042 0.156
(-1.140) (-1.616) (-2.862)
-0.710*** -0.710*** -1.031***
1996 0.789 0.794 0.929
(-11.789) (-11.789) (-22.593)
-0.225*** -0.320 -0.856***
1997 0.197 0.013 0.685
(-3.214) (-1.229) (-9.255)
-0.222*** -0.317 -0.527***
1998 0.223 0.036 0.371
(-3.448) (-1.550) (-4.903)
-0.549*** -0.445* -0.679***
1999 0.727 0.057 0.443
(-10.373) (-1.852) (-5.729)
-0.062 -0.305 -0.361***
2000 0.031 0.010 0.104
(-1.510) (-1.193) (-2.380)
-0.169 -0.0306 -0.403***
2001 0.035 0.020 0.084
(-1.556) ()-1.342 (-2.159)
-0.316*** -0.519** -0.390***
2002 0.262 0.073 0.150
(-3.851) (-2.015) (-2.805)
-0.428*** -0.640*** -1.022***
2003 0.239 0.181 0.439
(-3.639) (-3.100) (-5.609)
-0.123*** -6.097*** -1.861***
2004 0.453 0.140 0.690
(-5.629) (-2.647) (-9.134)
-0.213*** -0.391* -0.417***
2005 0.163 0.048 0.171
(-2.834) (-1.681) (-2.903)
Average -0.333 -0.916 -0.804
PS: the values in parentheses are t-value; *, ** and *** represent the significant level 0.1, 0.05 and 0.01, respectively.

Figure 1: The Power law exponents of performance measures

1
0
-11990 1992 1994 1996 1998 2000 2002 2004 2006

-2
-3
-4
-5
Year

ROA OITS NAVT


135 International Research Journal of Finance and Economics - Issue 46 (2010)
Figure 2: The Power law exponents of performance growth rates

0
-11990 1992 1994 1996 1998 2000 2002 2004 2006
-2
-3
-4
-5
-6
-7

Growth rate of ROA Growth rate of OITS Growth rate of NAVT

In order to examine the time trends of Power law exponents, we regress the exponents against
calendar time. The following equation expresses this regression.
qt = α + βt (Model 4)
Where α is the constant term and β is the coefficient of variable t . We presume 1992 as the
base year, so t =1 represents year 1992, and t =2 represents year 1993, and so on. If β is not
significantly differently from zero, we can know that the Power law exponent is stable over time. If
not, we can infer the qt is increasing or decreasing over time by observing the sign of β .
Table 3 shows the empirical results of equation 4. The coefficients ( β ) of ROA, growth rate of
ROA and NAVT are significantly positive, the coefficients are 0.073, 0.022, and 0.019 respectively.
These positive coefficients represent that the absolute values of the exponents of these three variables
are decreasing over time. The differences of these measures between banks decreased gradually.
Hence, these variables are more and more discriminating when evaluating bank’s performance. In
summary, we find that ROA/growth rate of ROA are more discriminating than other performance
(growth) measures when evaluating bank’s performance/growth. In addition, ROA is more
discriminating as a performance measure over time.

Table 3: The trends of Power law exponents

constant Coefficient( β ) adjusted r-square


-146.976*** 0.073***
ROA 0.558
(-4.185) (4.171)
-44.269* 0.022*
the growth rate of ROA 0.145
(-1.805) (1.792)
116.457 -0.059
OITS -0.031
(0.772) (-0.777)
2945.646 -1.476
the growth rate of OITS 0.064
(1.374) (-1.376)
-38.844** 0.019**
NAVT 0.343
(-2.862) (2.793)
-25.612 0.012
The growth rate of NAVT -0.067
(-0.444) (0.43)
ps: *, ** and *** represent the significant level 0.1, 0.05 and 0.01 respectively.
International Research Journal of Finance and Economics - Issue 46 (2010) 136

5. Conduction and Suggestion


Fundamental purpose behind measures is to improve performance. A good performance measure can
help organization to know where to improve, where to allocate or re-allocate money and people, and
which programs, methods, or employees are effective and efficient. This paper first takes advantage of
Power law to examine the distributions of measures and point out which measures are more
discriminating. Comparing with other financial measures, our empirical results indicate that ROA is
more discriminating. In addition, ROA is more discriminating over time. Hence, ROA is a good
performance measures.
There are two streams of performance indicators employed in prior researches. They are
accounting performance indicators and market performance indicators. In addition, the accounting
performance refers to the profitability and cost measures. This paper only focuses on three profitability
measures and their growth rates. For future extension, we can estimate the exponents of various
measures and find the discriminating indicator.

References
[1] Alegria, Carlos; Schaeck,2008, Klaus On measuring concentration in banking systems Finance
Research Letters ,5, Issue: 1, 59-67.
[2] Adrian Dragulescu, Victor M. Yakovenko, 2001,Exponential and power-law probability
distributions of wealth and income in the United Kingdom and the United States, Statistical
Mechanics and its Applications, 299, Issues 1-2 , 213–221.
[3] Axtell, R.L., 2001. Zipf distribution of U.S. firm sizes. Science 293 (5536), 1818–1820.
[4] Chen, Yanguang; Zhou, YixingMulti-fractal measures of city-size distributions based on the
three-parameter Zipf model Chaos, Solitons and Fractals Volume: 22, Issue: 4, November,
2004, pp. 793-805.
[5] Edvinsson, L. & Malone, M. S. 1997. Intellectual Capital: Realizing Your Company’s True
Value by finding it’s Hidden Roots. New York: HarperCollins Publishers Inc.
[6] E.W. Montroll, M.F. Shlesinger, J. Stat. Phys. 32 (1983) 209.
[7] H. Aoyama, W. Souma, Y. Nagahara, M.P. Okazaki, H. Takayasu, M. Takayasu, Fractals 8
(2000) 293.
[8] K. Okuyama, M. Takayasu, H. Takayasu, Physica A 269 (1999) 125.
[9] Katsuhiro Mizuno, Tetsuya Tsuji, Akio Kimura, Meigen Liu, Yoshihisa Masakado, Naoichi
Chino,2004,Twenty-seven years of complication-free life with clean intermittent self-
catheterization in a patient with spinal cord injury: A case report, Archives of Physical
Medicine and Rehabilitation, 85, Issue 10, 1705-1707.
[10] Lee, C.L. (2006), “Relationships among uncertainty, top management pay gap and firm
performance: a tale of tournament theory”, The International Journal of Accounting Studies, 42,
23-53. (In Chinese)
[11] L. Benguigui, E. Blumenfeld-Lieberthal, Comput. Environ. Urban (2007),
doi:10.1016/j.compenvurbsys.2006.11.002.
[12] Lucien Benguigui, Efrat Blumenfeld-Lieberthal ,2007,A dynamic model for city size
distribution beyond Zipf 's law, Statistical Mechanics and its Applications, 384, Issue 2, 15,
613-627.
[13] M. Naldi,2003, C oncentration indices and Zipf’s law, Economics Letters 78 (2003) 329–334.
[14] Maarten Bosker, Steven Brakman, Harry Garretsen, Marc Schramm, 2008, A century of
shocks: The evolution of the German city size distribution 1925–1999, Regional Science and
Urban Economics,38, Issue 4,330-347.
[15] M.H.R. Stanley, L.A.N. Amaral, S.V. Buldyrev, S.V. Havlin, H. Leschhron, P. Maass, M.A.
Salinger,H.E. Stanley, Nature 397 (1996) 804.
[16] Pareto, V., 1897. Cours d’E´ conomie Politique, F. Rouge, Lausanne.
137 International Research Journal of Finance and Economics - Issue 46 (2010)

[17] P.V. (Sundar) Balakrishnan, James M. Miller, S. Gowri Shankar,2008, Power law and
evolutionary trends in stock markets, Economics Letters,98, Issue 2, 194-200.
[18] Ramsden, J.J.; Kiss-Haypál, Gy. Company size distribution in different countries Physica A
Volume: 277, Issue: 1-2, 220-227.
[19] R. Mansilla, E. Köppen, G. Cocho, P. Miramontes,2007, On the behavior of journal impact
factor rank-order distribution, Journal of Informetrics, 1, Issue 2, 155-160.
[20] Stanley, Michael H.R.; Buldyrev, Sergey V.; Havlin, Shlomo; Mantegna, Rosario N.; Salinger,
Michael A.; et. al. Zipf plots and the size distribution of firms Economics Letters Volume: 49,
Issue: 4, October, 1995, pp. 453-457.
[21] S.V. Buldyrev, L.A.N. Amaral, S. Havlin, H. Leschhorn, P. Maass, M.A. Salinger, H.E.
Stanley,M.H.R. Stanley, J. Phys. (France) I7 (1997) 635.
[22] Shi-Ming Huang , David C. Yen , Luen-Wei Yang , Jing-Shiuan Hua,2008. An investigation of
Zipf's Law for fraud detection, Decision Support Systems.
[23] Xavier Gabaix, Parameswaran Gopikrishnan, Vasiliki Plerou, H. Eugene Stanley,2003,
Understanding the cubic and half-cubic laws of financial fluctuations, Statistical Mechanics and
its Applications, 324, Issues 1-2, 1 – 5.
[24] Zipf, G.K., 1949. Human Behavior and the Principle of Least Effort: an Introduction To Human
Ecology, Addison-Wesley.