One may wonder what derives the high performance of the “neighbouring assets” strategy. In this paper, I show that if we classify assets into 10 portfolios based on the past performance of their neighbours and create hedge portfolios, the long-short portfolio easily generates strong and significant alphas, both statistically and economically.
Now, the question here is: “which characteristics” derive this performance? But wait! Here is exactly the main point of my paper: characteristics jointly derive this performance.
So, asking the question of “which characteristics” is not really that accurate. For instance, if I consider size in isolation, there’s literally no profitability in this strategy. But if I add size to a set of other characteristics, the spread between portfolios starts to appear.
In one table, I show that if we add more characteristics, the performance starts increasing. Does the performance increase because of the new characteristics? Not necessarily. But it increases because of the interactions between new characteristics and the previous ones.
How the interaction between characteristics should enter the model? For instance, should we consider size×bm, or size2×bm or size2×bm2 or size2×bm2×momentum …? We don’t really know the answer. And we are subject to the curse of dimensionality. Simple interactions might have been working decades ago but probably investors learned the mispricing and traded against them, which makes the relationship between characteristics and expected returns more complicated.
Here is the beauty of “Neighbouring Assets” that we let data speak. The neighbouring assets will tell you about this relationship without providing a closed-form solution.
Leave a Reply