A reader commented and asked: “I am interested in Mebane Faber’s 10-month SMA timing strategy, as it seems to match the market with less risk and outperform other moving average strategies I’ve seen. Based on the results of ‘Is There a Best SMA Calculation Interval for Long-term Crossing Signals?’, it seems that Faber’s strategy is not brittle as far as choosing an 8-, 10-, 12- or 14-month SMA. However, what if I were to trade on a day other than the end of the month? Would I get drastically different results? If so, that might suggest that Faber’s choice of day is ‘data mining’ and the performance of his strategy may not persist.”
While not precisely addressing your question, the following points are perhaps close enough in concept to address the concern.
The momentum investing approach, whether implemented via simple moving average (SMA) crossovers or lagged returns (or a combination) may be fragile with respect to exact parameter selection:
- The second chart and associated discussion of 200-day SMA crossovers in “The TimingCube Market Timing Advisory Service” finds that results for a 200-day SMA crossover strategy are sensitive to anticipating the crossover (trading at the same close as the crossover). Waiting a day after crossover substantially reduces crossover strategy returns.
- As found in “Simple Sector ETF Momentum Strategy Robustness/Sensitivity Tests”, there is substantial variation in outcomes with the momentum ranking interval, making the six-month and ten-month ranking intervals look lucky.
- A reader reported in response to this second analysis his own findings that shifting the momentum ranking interval off the end of the month (for a six-month ranking interval) affects returns substantially, making the end of the month appear lucky. This kind of testing is very time-consuming and may involve some judgments about what constitutes a month.
Possible explanations for such fragilities include:
- The specific strategy parameter values are lucky, and luck will not persist outside the test sample.
- Many investors have adopted and are acting on these parameter values, and results may persist as long as investors continue to act synchronously.
- The sample periods in the tests are too short (or the underlying distributions are too intractable) for reliable inference.
It seems plausible that the end-of-the-month choice for momentum ranking and the anticipatory execution of 200-day SMA crossover signals are “natural selections” rather than data snooping artifacts. The reason for choice of momentum ranking interval length seems more arguable.
Mal Williams reported results for a sensitivity test of his momentum-based asset class rotation strategy (see “An Investor’s Asset Class Momentum Trading Strategy”). He tested trading 1, 15, 30 and 60 days after the initial signal (requiring that the asset still be on the buy list after the delay) — the longer the delay, the lower the overall portfolio return over a 20-year period. While total return is higher for trading right away, so is volatility (but risk-adjusted return is still higher). “Perhaps the magic is being early in responding to a signal.”
