A cognitive bias of traders in the stock market

Abstract

In this paper, I investigate whether investors in the stock market use round numbers as cognitive reference points during the trading process. Using transaction-level data provided by Abel Noser, I confirm that even for professional institutional investors, their transaction prices tend to cluster at round-number prices. And the degree of the clustering is consistent with the accessibility of the numbers, i.e. whole dollars, half dollars, dimes and nickels, which I refer to as the 'round-number' bias. Then, within the Abel Noser sample, there is a considerable heterogeneity of the 'round-number' bias: those trades submitted by institutions in larger size groups or those that have specialized trading departments exhibit lesser degree of the bias, comparing with trades from smaller institutions or do not have designated traders inside. Also, building on some findings in prior studies using the same data set, I find that at broker level, those trades executed by 'discount' brokers (who mainly focus on executing trades on behalf of their clients and charge lower commissions) are less clustered at round numbers, comparing with those executed by traditional brokerage houses. These cross-sectional differences indicate that the degree of the bias is related to the efforts devoted by investment companies and brokers to the trade execution process. Finally, to further study the plausible causes of the 'round-number' bias during the trading stage, I move on to the TAQ data set, which contains most of intraday transactions for securities listed in major stock exchanges in the U.S. Using different weighting methods, I find that on an average trading day, the degree of the 'round-number' bias in the overall market is much higher comparing with that in the Abel Noser sample, suggesting that retail investors are severely affected by such price preference when submitting their orders. And due to the over clustering around 'round number' prices, those 'round-number' trades incur higher trading costs (measured by the commonly used 'effective spread'), ranging from an annual amount of 200 to 900 million of dollars during our sample period (2001-2014). Yet, we do observe a significant declining trend of the 'round-number' bias across time, which can be attributed to the speedy and broad adoption of the algorithm-based trading practice.