In 2007, The Journal of Wealth Management published a paper by Mebane Faber of Cambria Investment Management, “A Quantitative Approach to Tactical Asset Allocation.” The paper reviewed the results of a simple market-timing strategy using a 10-month version of the popular 200-day moving average technical indicator (in other words, a time-series momentum strategy). It went on to become the most downloaded paper on SSRN’s website, with more than 200,000 downloads.
Ten years later, Faber thought it was time to review how the strategy had held up out-of-sample. The update, “A Quantitative Approach to Tactical Asset Allocation Revisited 10 Years Later,” was published in a special 2018 issue of The Journal of Portfolio Management.
In his update, Faber noted: “It’s important to understand that ‘beating the market’ was never the goal of the model. The intent was to identify a trading system that largely approximated market returns, yet did so with significantly less volatility. The reason for this was simple—emotions can wreak havoc on investors’ ability to follow their stated investment plan. All too often, we fall victim to fear when markets have turned against us and sell at nearly the worst possible time.”
The original article was published with data up through 2005, so Faber examined the historical in-sample results (1972 through 2005) as well as the out-of-sample returns in the 11 years since. He found that from 2006 through 2016, the simple timing strategy had provided higher returns than the S&P 500 Index (8.5% versus 7.7%), did so with much less volatility (7.4% versus 14.7%) and had a much lower maximum drawdown (16.7% versus 51%). The result was a dramatic improvement in Sharpe ratio, from 0.45 to 0.80.
In reviewing his findings, Faber noted: “Though the timing system outperformed by a
significant amount during the 2008–2009 bear market, it went on to underperform stocks six of the next eight years. Many investors who had implemented the timing model after the crash likely struggled with staying the course with a tactical approach.”
Pervasiveness Of The Data
One test of whether investors should have faith in the results of an investment strategy is that it also work in other asset classes—reducing the risk of data mining. Thus, in addition to reviewing the results in U.S. stocks, Faber also reviewed the performance of his 10-month moving average strategy across foreign stocks, U.S. bonds, REITs and commodities. The timing model used equal weightings and treated each asset class independently; it is either long the given asset class or its 20% allocation sits in cash.
This second analysis produced the same result: improved returns (4.9% versus 3.5%), much lower volatility (6.6% versus 12.8%), dramatically higher Sharpe ratio (0.59 versus 0.19) and much lower maximum drawdown (9.5% versus 46%).
Robustness Of The Data
Another important test of whether investors should have faith in an investment strategy is that it be robust to various definitions. Faber found similar results for strategies using periods of six, eight and 12 months.
He also tested a more conservative strategy, which had a 40% allocation to U.S. bonds, and a more aggressive strategy, which begins with the moderate allocation and then selects the top three of the five assets he examined as ranked by an average of one-, three-, six- and 12-month total returns (time-series momentum), only including assets if they are above their long-term moving average (otherwise that portion of the portfolio is moved to cash using T-bills).
In addition, Faber tested the strategy using long-term bonds for cash management. Finally, he tested a more diversified version of the strategy, using a long list of other asset classes, including TIPS, high-yield bonds, emerging market bonds, foreign REITs, fundamental indexes and currencies.
The results were similar in each of the tests. And with the help of diversification benefits, the broader strategy, which included more asset classes, improved results, increasing the return of the quantitative strategy from 9.8% to 11.3%, with only a modest 0.2% increase in volatility. The result was that the Sharpe ratio increased from 0.71 to 0.91.