Night Trading: Lower Risk But Higher Returns?VW Staff
Night Trading: Lower Risk But Higher Returns?
San Diego State University – Finance Department
July 18, 2015
This paper demonstrates that overnight returns are subject to highly persistent biases and examines the profitability of overnight-only investments in that context. Overnight returns tend to exceed their intraday counterparts, and the paper first reconciles these patterns by introducing a model that factors in recurring biases. This model identifies one fifth of stocks as having positive and statistically significant overnight biases. Investing overnight in these stocks in the next year yields twice the market’s return for a third of the market beta. Results have also implications for daytime investors as these stocks average negative returns intraday. Implementation costs and issues are discussed.
Night Trading: Lower Risk But Higher Returns? – Introduction
Most investors are unaware that overnight returns exceed intraday returns in the United States, although this phenomenon is documented in several recent works (Cliff et al. (2008); Kelly and Clark (2011); Berkman et al. (2012)). This observation is surprising given that it is well-known that volatility is lower during the nighttime period. This paper contributes to the literature in four ways by providing new evidence that the overnight return anomaly is pervasive, persistent, predictable, and can be profitable after costs. The paper broadens the scope of the effect by showing that overnight returns beat intraday ones in each of the 23 countries of MSCI’s World Index. Moreover, in each country, sorting stocks by past overnight returns remarkably and consistently predicts future overnight returns: A strategy as simple as buying stocks from the previous year’s top quartile averages an annual overnight return of 43.7% the next year. This anomaly deserves more attention as it is not limited to a period or country and its superior performance persists several years after portfolios are formed.
More specifically, the paper examines two research questions related to overnight returns. First, is there long-term persistency in overnight returns? In regression analysis, the paper finds that the past year’s overnight return is actually the most important predictor of future overnight returns, followed by volatility, turnover, and momentum.1 Second, can these patterns be exploited profitably and reliably with overnight-only investment strategies, even after risk and costs are taken into account? The cost of overnight strategies is generally not quantified in current works, which limits their explicit recommendations for the timing of existing trades— i.e., delaying purchases to the close and sales to the open. Thus, analyzing costs fills a gap in this literature and expands applications to stand-alone alpha-generating investment strategies, which entail larger profit opportunities.
Overnight strategies would be cost-prohibitive with round-trip transaction costs such as those reported in Barber and Odean (2001): 1% for the bid-ask spread and 1.4% in commissions. However, transaction costs have declined sharply over the last decade, and the paper calibrates commission rates realistically by surveying broker fee structures for 2014 in the United States and finding the lowest costs. From a cost standpoint, overnight strategies offer an interesting case because they can be implemented with the exchanges’ single-price opening and closing call auctions, saving on the bid-ask spread. The paper finds that the resulting average daily round-trip costs are between 1.5 and 3.1 basis points, depending on the size of the trades.
Hence, abnormal overnight returns must be at least 4%-8% annually to offset transaction costs. Overnight investment strategies should aim to identify stocks that are likely to exceed that threshold by a comfortable margin. The paper’s finding that stocks exhibit persistency in their overnight returns facilitates this identification. This is demonstrated formally with Fama-MacBeth (1973) cross-sectional regressions and two datasets. The first sample is based on data from the Center for Research in Security Prices (CRSP) covering the period 1995-2014 for the United States. The second sample uses the constituents of major indexes representing 22 developed countries over the period 2002-2014. With both samples, the regressions show that overnight monthly returns are strongly positively related with each of their lagged values, with lags ranging from one to 60 months. By contrast, the relationship of overnight returns with current and past monthly intraday returns is negative.
See full PDF below.