Daily and Intraday Herding within Different Types of Investors in Borsa Istanbul

ABSTRACT

This paper aims to explore the daily and intraday herd behavior of various investor groups trading in an emerging equity market, Borsa Istanbul (BIST). We analyze a one-year tick-by-tick order and trade data of BIST 100 Index stocks and document differences in herding behavior of investor groups considering market capitalization, market conditions, and announcements as well as daily and intraday periodicities. We find that nonprofessional investors (brokerage houses and domestic funds) tend to herd on large (small) stocks; their herding behavior mostly exhibits a U shape (an inverse U shape) during the day. All types of investors tend to herd in down markets on a daily basis while this behavior disappears, even inverts intraday.

KEY WORDS:

• behavioral finance
• Borsa Istanbul
• herding

JEL:

• G41
• G23
• G12

Notes

1. Alternatively, we constructed five sub-periods of 35 (39) minutes in the morning (afternoon) and four sub-periods of 43.75 (48.75) minutes in the morning and afternoon (thanks to an anonymous referee for this point). Results are qualitatively same. For the sake of brevity, we do not report these results.

2. This group includes any individual or corporate customer, both domestic and foreign, having an investment account.

3. Any proprietary trading activity of brokerage houses.

4. These involve mutual funds, pension funds, ETFs as well as hedge funds.

5. http://www.tuyid.org/files/yayinlar/BIST_Trends_Report_VII.pdf http://www.tuyid.org/files/yayinlar/BIST_Trends_Report_IX.pdf.

6. Currently, around 34%.

7. All trading values are expressed in logarithms. In order to avoid any negative or infinite numbers due to the nature of logarithm, we add 1 to any trading value in the dataset.

8. See, e.g. Odders-White (2000) and Chakrabarty, Moulton, and Shkilko (2012) for evidence on weak classification performance of algorithms. Aktas and Kryzanowski (2014) and Ersan and Alıcı (2016) apply the same methodology in their BIST analyses.

9. ±1% return levels can be assumed as extreme movements in some stock markets, especially in developed ones. For instance, from June 3, 2013 to May 30, 2014, only 15% of daily returns in S&P500 are below or above ±1% cut-off point. However, ±3% (±1% to ±3%) cut-off points for the extreme (significant) daily movements in BIST 100 Index seem appropriate as round cut-off values since a return between ±1% to ±3% form approximately 45% of our sample period. At least ±3% daily variation in BIST 100 Index return covers almost 11% of our sample time interval. As a robustness check, we further analyze the impact of up and down market movements on trading value dispersion with ±4% (±1% to ±4%) cut-off points as well as z-scores with σ (2σ), for the extreme (significant) daily movements in BIST 100 Index. The results are similar.

10. We replicate our analyses with regular dummy variables assigning one on days with an absolute return of 1% to 3% (and more than 3%) absolute return and zero otherwise. The results are identical.

11. Two intervals exactly account for 44.9% and 11.1% in total number of sub-periods. We replicate our analysis with round cut-off values of 0.15% and 0.5% for significant absolute return and 0.5% for extreme absolute return. Additionally, we employ cut-off values with z-scores (σ and 2σ). Results are qualitatively identical.

12. We replicate our analysis with a dummy variable taking the value of one if there exists a firm-level news announcement on a given day or intraday period and zero otherwise. The results are qualitatively same.

13. This variable is formed as a dummy variable by its nature; we observe a maximum of one (or very rarely two) macroeconomic news release per day or per intraday interval covering our sample period.

14. Following Cooper, Gutierrez, and Hameed (2004), we define the market state as bearish (bullish) when the market’s six-month return is negative (positive). See, among others, Walter and Moritz Weber (2006), Arouri et al. (2013) and Rubbaniy (2016) for herding analyses considering bull/bear market conditions.

15. For the sake of brevity, results are not included in the paper.

16. Significant differences in Table 3 result from the relatively low variability in our group-based trading value dispersion levels. For example, for the ‘all stocks’ portfolio, mean daily cross-sectional dispersion difference between domestic funds and brokerage houses is −0.42, the lowest absolute difference in the table. This difference between DF and B groups is significant at 1% level since mean dispersion values for 248 trading days are 6.54 and 6.12 with low standard deviations of 0.24 and 0.23, respectively. DF dispersion is higher than B dispersion in only 13 days out of 248.

17. Our claim on the larger role of firm-specific announcements on herding is driven by the negative coefficient of daily firm-specific news variable in the regression for the IC group. This group constitutes the majority of the trading activity in our dataset (89%). Observation of larger extent of herding by this group on the days with firm-specific news but not on the days with macroeconomic news is interesting and informative about the Turkish market.

18. We find this result in spite of a limited share of foreign traders within the IC group. This finding would be amplified if individual investors could be fully separated.

Additional information

Funding

We are grateful for the financial support provided by the Scientific and Technological Research Council of Turkey (TUBITAK) for the Project No [117K908].