Time-series momentum is a very venerable strategy, especially in the commodities world. By virtue of a meaningless coincidence, “venerable” sounds a lot like “vulnerable.” And, fittingly, the TSMOM strategy is vulnerable to sudden shifts, such as that created by the global financial crisis of 2008.
Let’s start with definition. A momentum strategy without adjective typically uses the performance of an asset against its peers in the recent past to predict performance in the near future. Time-series momentum on the other hand says, “Forget the peers.” It looks only to the recent performance of an asset considered in isolation.
In either case, the momentum strategy sounds a lot on the surface like the strategy of a gambler who says “black has been coming up a lot recently, I’ll bet on black,” as opposed to the other fellow at the table who says, “black has been coming up a lot lately, red must be due.” Casinos will tell you that both the momentum bettor and the contrarian are perfectly welcome to lose their money at the same table. And, unfortunately, the analogy continues to seem appropriate once one scratches beneath the surface.
A New Paper
In a new paper, Nick Baltas of UBS Investment Bank and Robert Kosowski of the Imperial College Business School look at several ways of improving the performance of the time-series momentum strategy, beginning with the observation that it has performed quite poorly in the post 2008 period. These same two authors, in a 2013 paper, had found that there was no significant evidence of capacity constraints in the performance of the strategy, and they proposed that its post-crisis problems were due to (1) the lack of significant price trends, and (2) the increased level of correlation across assets of different asset classes from 2008 to 2013.
From these two postulates, in their new paper, they move to a theorem. Performance of TSMOM strategies has been bad because the crisis generated a lot of pair-wise trading, and the climate since then has been very bad for pairwise trading.
These authors suggest, then, that “incorporating information from the correlation matrix of the assets into portfolio construction can render the [TSMOM] strategy more robust in periods of increased co-movement.” More specifically, trading institutions might use the contemporaneous level of average pairwise correlations to adjust their target vol level for each asset to which this strategy is applicable.
The paper, “Demystifying Time-Series Momentum Strategies,” also addresses the issue of transaction costs in the context of TSMOM. Assuming a constant level of market impact, the authors infer that such costs will be proportional to portfolio turnover. Intuitively (that is, non-mathematically) one senses a problem: won’t the adjustments for which they have just called in the proceeding section, those re-jiggerings of the target volatility level because of changes in co-movement, increase the turnover and thus the transaction costs?
Well, yes, they acknowledge, that is exactly what will happen. But the benefits of the extra rebalancing survive factoring in those higher costs.
The paper, considered on its own terms, raises a number of questions. For example, given that transaction costs are the product of market impact and portfolio turnover, why would one want to simplify that already very simple formulation further by assuming again that market impact remains constant? Doesn’t this bear on the issue of capacity constraint for the strategy? If market impact increases with increasing order size, then at what level of order size does the rebalancing cease to be worthwhile?
To their credit the authors acknowledge the tentative nature of their inferences. Scholars studying analogous questions in the world of equities have an easier time of it, they say, since they can use institutional equity order data, which are not available for the futures contracts studied in the Baltas-Kosowski paper. They do issue the customary scholarly call for further research on the issues they leave open.
The authors also offer advice on how institutions might go about estimating volatility and price trends more efficiently, the better to ride their presumed momentum.
Old wisdom is best though: stay away from the tables unless you have an edge.