Do long-term stock market timing models work? If so, which type works best? In their October 2005 paper entitled Timing is Everything: A Comparison and Evaluation of Market Timing Strategies, Chris Brooks, Apostolos Katsaris and Gita Persand investigate the profitability of several timing models over a very long sample of S&P 500 index returns. Specifically, they test the timing power of: (1) the ratio of the long-term Treasury bond yield to the stock dividend yield; (2) the spreads between the stock earnings yield and the yields on either the three-month Treasury bills (T-bills) or the 10-year Treasury notes (T-notes); (3) a model for predicting when bear markets will occur based on the spread between T-note and T-bill yields; and, (4) an approach for predicting market turning points based on speculative bubbles. Timing signals trigger binary switching between stocks and T-bills. Using monthly stock return and model parameter data from January 1871-December 1926 for initial model calibration and January 1927-August 2003 for model testing and recalibration (a total of 1,592 months), they find that:
- Before trading frictions, the model based on the difference between the stock earnings yield and the T-bill yield generates the highest average raw return, 0.96% per month (12.1% annualized), versus 0.94% for a buy-and-hold benchmark.
- All of the timing models have lower standard deviations of returns than buy-and-hold, offering Sharpe ratios (before trading frictions) from 0.11 for the speculative bubble model to 0.14 for the model based on the difference between the stock earnings yield and the T-bill yield (compared to 0.12 for buy-and-hold).
- Over the 80-year test period, the number of round trip (buy-sell) transactions ranges from just five for the model based on the difference between the stock earnings yield and the T-note yield to 24 for the speculative bubble model.
- Empirically optimizing parameters of the tested models boosts their Sharpe ratios by 10%-40%.
The following table, extracted from the paper, summarizes average monthly performance during January 1927-August 2003 before trading frictions for all tested models using non-optimized parameter values. Model abbreviations are:
- GEYR (“gilt-equity yield ratio”)- the ratio of the long-term Treasury bond yield to the stock dividend yield.
- EPS – the spread between the stock earnings yield and the T-bill yield.
- EPL – the spread between the stock earnings yield and the T-note yield.
- Bear – the model for predicting when bear markets will occur based on the spread between T-note and T-bill yields.
- Bubble – an approach for predicting market turning points based on speculative bubbles.
Two timing models (EPS and Bear) edge out buy-and-hold based on raw average monthly returns. Four of the five timing models beat buy-and-hold based on Sharpe ratios. None of the models trade frequently. The worst-performing model based on raw average return and Sharpe ratio (Bubble) spends the least time in equities. However, GEYR and Bubble (with the least exposure to equities) are the top performers based on more complex assessment of the return distribution (omega ratio).
In summary, long-term stock market timing models may enhance investment returns, especially on a risk-adjusted basis. Which model is best depends on the risk-adjustment metric used.
