Herding behaviour and the declining value relevance of accounting information: evidence from an emerging stock market

ABSTRACT

This article sheds light on the underlying mechanisms behind the changes in the value relevance of accounting information in the Karachi Stock Exchange (KSE) during the 1999–2010 period. We find that neither changes in earnings quality nor the earnings lack of timeliness hypothesis can explain the decline in the value relevance of accounting information in the KSE. Based on the stylized facts associated with the growth of the KSE and the broader economics literature, we argue that the reduction in the explanatory power of accounting information vis-à-vis stock returns was caused by herding behaviour. Empirical estimates from state-space model of herding behaviour confirm the existence of herding, and we find that the value relevance of accounting information is significantly lower in periods characterized by herding behaviour. This article is also amongst the first attempts to empirically demonstrate that an expansionary monetary policy and increases in foreign portfolio investment lead to increased levels of herding.

KEYWORDS:

• Herding behaviour
• value relevance of accounting information
• emerging markets
• KSE

JEL CLASSIFICATION:

• G12
• G15
• M41

Acknowledgement

We would like to thank Mark Taylor, the Editor, and an anonymous reviewer for providing valuable comments that helped us improve the quality of this article

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ In this article, noise traders refer to investors who use unsophisticated methods, usually based on market sentiment, to formulate their investment strategies and often ignore information on firm fundamentals.

² Asset pricing models (e.g. CAPM) are used to pin down a stock’s intrinsic value because intrinsic values are also unobservable. Thus, an empirical test of changes in levels of noise trading is also a joint test of the underlying asset pricing model. Unfortunately, there is no solution to this ‘joint-hypothesis problem’ in the literature (Fama 1991).

³ Teoh. and Wong. (1993) find that a higher percentage of audits by Big-4 accounting firms’ leads to higher value relevance of financial statements because audits by Big-4 firms improve the credibility of financial statements in the eyes of investors. Roughly 65% of the firms in our sample were audited by one of the Big-4 accounting firms (i.e. EY, KPMG, PWC and Deloitte).

⁴ The original model includes growth in investment and lagged earnings yield as additional proxies for measurement errors. However, coefficients on both variables were found to be statistically insignificant and hence are not included in our reported model.

⁵ Mean number of forecasts per firm varied from a low of 3.24 in Pakistan to 9.23 in Philippines and 10.67 in Malaysia. Similarly, mean forecast dispersion in Pakistan was significantly higher compared to India and Malaysia.

⁶ The term ‘quality’ of stock prices refers to the ability of stock prices to impound information on fundamentals.

⁷ Pi and Timme (1993) find that the sophisticated investment criterion and long-term investment horizons of institutional investors positively affect the quality of stock prices. Hence, a dearth of institutional investors increases the likelihood of persistent mispricing.

⁸ If ${\sigma }_{mn}^2=0$, the measurement equation becomes $log\left[St{d}_c\left({\beta }_{imt}\right)\right]={\mu }_m+{\upsilon }_{mt}$, indicating that herding behaviour is not present. The magnitude of estimates of ${\varnothing }_m$ should be less than 1 to ensure that the process does not explode.

⁹ Pakistan descended into a period of great political instability after former Prime Minister Benazir Bhutto was assassinated in December 2007. At the same time, surging oil prices and a slowdown of the global economy in the aftermath of the 2008 financial crisis also contributed towards the crash of KSE.

¹⁰ If we interpolate R-square from value-relevance regressions over monthly frequency, then our dependent variable will have no variation over 12 consecutive data points. Clearly, estimates based on such a data set will be biased towards insignificance.

¹¹ The approach of utilizing a dummy variable in a single model is more robust compared to splitting the sample and estimating the model separately for herding and non-herding years.

¹² Test of ${\beta }_1={\bar{\beta }}_1$ in the pooled OLS model yielded a Wald test-statistic of 16.45 (p-value <0.0001). Test of ${\beta }_1={\bar{\beta }}_1$ in the fixed-effects model yielded a Wald test-statistic of 12.97 (p-value <0.0003).

¹³ Results (not reported) from regressions with adjusted R-square as the dependent variable and average herding levels as the independent variable along with controls for earnings quality lead us to the same conclusion.

¹⁴ We re-estimated the model with other macroeconomic variables, e.g. remittances, exchange rates and T-bill rates, but none had any statistically significant impact on herding behaviour.