Welcome to the second issue of the fifth volume of The Journal of Investment Strategies. In this issue, we present four papers on the topics of defining and testing investment strategy predictors (alphas), a method of statistical stock selection, and a methodology for optimizing trading and execution.
In the first paper of the issue, Adam Zaremba and Przemysław Konieczka explore a quantitative country selection strategy in the context of international stock investments using investable country-specific indexes across a large set of both developed and emerging economies. The authors test standard stock-selection strategies including those based on size, value, quality, momentum and volatility. While they find that size and leverage, as measured across markets, do lead to outperformance, the rest of the metrics appear to have insignificant or unreliable predictability. The authors also show that the capital market constraints further weaken the excess returns. The lesson that I take from this paper is that cross-country selection using aggregate country indexes is not a useful strategy, perhaps in the same way that one can rarely make a reliable quantitative strategy by allocating across industry sectors based on common metrics. The industry and country groupings are indeed very important in explaining the variability of returns, but they do not necessarily help to forecast them. One really has to select stocks, not countries or sectors, and when doing so I amsure it is possible to earn a decent excess return, while taking into account country specifics. There are good and bad investments everywhere, and lumping them together simply makes the portfolio manager's job too difficult.
In our second paper, Libin Yang, William Rea and Alethea Rea present an interesting way to approach stock selection and portfolio construction using purely statistical methods. The paper follows the advent of other risk-driven portfolio construction techniques, including minimum volatility, risk parity and others. The paper takes the view that optimizing the diversity of the portfolio is the objective, and the authors show how one can achieve that objective given a target portfolio count: the number of different stocks allowed in the portfolio. It can be considered an alternative to other maximum diversity methodologies. The method is shown to work well for the Australian Stock Market index, allowing it to be replicated with as few as fifteen stocks. Furthermore, the paper clarifies that a target size of the replicating basket is not constant over time, and depends on market conditions, which is not surprising given that the manner in which we arrive at this is by perusing the covariance matrix of asset returns, which is clearly dependent on market conditions. While I am generally very sympathetic to risk-based portfolio methodologies, I find the objective of maximum diversification somewhat less straightforward than others in this area. Why would an investor want a highly diverse portfolio? Presumably because he or she wants to minimize the risk and increase the robustness of the portfolio returns. But if so, these objectives can be more directly addressed, such as in minimum volatility or risk parity. The definition of diversity is a somewhat roundabout way to get to the same results. To the extent that it is different, it must be that it provides some other dimension of robustness that the other approaches that are more firmly grounded in risk measures do not. I would not be surprised if it was indeed so, but I would need more direct evidence of such robustness before declaring my preferences.
In the issue's third paper, Gabriel H. Tucci and M. Valentina Vega derive optimal trading trajectories for executing large orders under linear and nonlinear market impact assumptions. The studies of dynamic optimization of trading strategies have taken center stage in recent years, partly due to the proliferation and eventual dominance of algorithmic trading in the modern markets. Such trading not only serves as a primary strategy for high-frequency-trading firms, but it also serves as a key plank of the execution efforts of the vast majority of other investors, whether such algorithms are implemented internally by large asset management firms or by brokers that cater to mid-size and other less specialized market participants. Regardless of whether the reason for the trade is fundamental, tactical or quantitative, the execution of such trades is very often algorithmic. Therefore, every study that discovers a specific way to implement such trading is bound to be relevant and full of market impact (pun intended). This paper is no exception, and I am sure that many execution desks will carefully study it to see if it offers a new edge in the endless game of shaving fractions of basis points off the cost of trading.
In the Investment Strategy Forum, we welcome a paper by Zura Kakushadze and Igor Tulchinsky, who share with us their insights from the analysis of literally thousands of "alphas", or, in other words, predictive indicators that are actually used by Tulchinsky's WorldQuant in real-life investment management. These indicators are all relatively short term in nature, ranging from about a day to just under a month in typical holding periods. As a result, they generate a fairly large turnover, and one might inquire whether such turnover by itself leads to value added, ie, to an increase in expected returns. The answer turns out to be no, the expected returns do not depend on the turnover, and the more frequently rebalanced strategies tend to make proportionally fewer cents per dollar for each rebalancing, and the return per unit of time therefore remains largely independent of the frequency. Why then would one go to the trouble of trading more frequently? I think it is mostly because there are many more ways to trade when you can trade often compared with when you hold positions longer. Thus one can construct many more "substrategies", or "alphas", with shorter horizon, and therefore eventually create greater diversity for the multi-strategy portfolio. In contrast with turnover, the return actually does depend on the volatility of the strategy, but this dependence is slightly less than linear, thus implying that strategies with greater volatility will have to endure lower Sharpe ratio - which is a rule of thumb familiar to many fund-of-funds managers. You can have high Sharpe ratio but it usually requires low volatility and thus one must lever it up to reach reasonable return targets - and therein lies the rub, because adding leverage also adds additional tail risks, thereby reducing the allure of these strategies and restoring, to some extent, the level playing field among the different types of trading strategies. I would like to thank our readers for their continued support and interest, and hope that they will find something useful in this issue of The Journal of Investment Strategies - and I promise that we will continue to offer equally interesting issues in the future.
Arthur M. Berd
General Quantitative LLC
This paper compares sixteen distinct country-selection strategies within a sample of seventy-eight countries between 1999-2015.
The authors of this paper propose a stock selection method based on a variable selection method used with PCA in multivariate statistics.
This paper derives explicit formulas for the optimal implementation shortfall trading curve with linear and nonlinear market impact.
This paper analyzes empirical data for 4000 real-life trading portfolios with holding periods of about 0.7-19 trading days.