Sina Seyfi

This figure shows my high-dimensional portfolio sort that makes the basis portfolios. My algorithm groups stocks into portfolios (e.g., deciles) such that stocks within each portfolio have the closest distance of characteristics, and the number of stocks in all portfolios is the same. In case there is only one characteristic, my method delivers exactly the common univariate sort in the literature.

Abstract: I propose creating a small set of well-diversified high-dimensional basis portfolios such that stocks within (across) portfolios have the most (least) similar fundamentals, proxied by a large set of characteristics. If the comovement between stocks is a function of a large set of characteristics, the high-dimensional basis portfolios that are distinct in all characteristics show low comovements and high dispersion in expected returns. As a result, the optimal portfolio spanned by high-dimensional basis portfolios displays a sizeable out-of-sample Sharpe ratio of 1.78 with a monthly alpha of 1.71% (t = 11.11), without taking any extreme position on any asset.

According to which characteristics do low-mean stocks differ from high-mean stocks? This figure shows how much of the distance between characteristics of losers from winner stocks comes from each characteristic. Accounting for 5.7% of the distance, momentum is the most important characteristic. Moreover, almost 15 characteristics constitute 50% of the distance.

Abstract: I develop a method to identify the strongest determinants of expected returns. Instead of sorting stocks on characteristics, I sort stocks into portfolios based on their realized returns—the variable of interest—at each month in the past and find the average of each characteristic among assets in each portfolio. Then I create out-of-sample portfolios such that they are as similar as possible to the returns-sorted portfolios regarding 178 characteristics. This approach separates low-mean stocks from the high-mean ones so that a long-short portfolio gains an out-of-sample monthly alpha of 1.74% (t = 13.78). Characteristics that differ between low- and high-mean stocks drive the dispersion in expected returns. I find price-based characteristics are the strongest predictors.

Without loss of generality, if there exist only three firm characteristics, the k closest assets to the asset j (shown as red dots) are defined as the neighbours of asset j. If expected returns are a function of firm characteristics, regardless of the functional form, neighbouring assets display similar expected returns.