Everyone who has been investing in the stock market for more than a week knows that stocks do not always go up. Downturns can be prolonged and painful. It is not uncommon to see your hard-earned capital lose 15%,20%, and even more, in a matter of a few months. Those of us who have been investing through the Great Recession of 2008-2009 have seen their capital lose more than half of its value.
But what if I told you that there is a way to enjoy the uptrends, yet avoid major downtrends?
This is what Market Timing is all about. At least what its proponents claim.
Market timers devise signals for entering the market (or stock portfolios) and exiting it in an attempt to enjoy stocks’ appreciation potential but keep out while the market declines.
Well, do those market timing techniques have real merit? Or is it just a pipe dream? In this article, we shall examine the data and develop a more educated opinion.
But before diving into figures and charts, let us clarify the definitions first. Market Timing is a practice of switching between a risk asset and a safe asset at favorable times. A Risk Asset – the quantitative strategy, stock, or ETF under test. The Risk Asset is volatile and prone to declines, hence it can benefit from Market Timing. A market benchmark, such as the S&P 500 or the Russell 2000, is a risk asset. A quantitative portfolio is a risk asset. So is an individual stock. A Safe Asset – Cash or short-term bonds. In the backtesting simulations described below, the risk assets are the quantitative strategies on our website, and for the safe asset, we chose cash. Had we used bonds, then the total returns of the portfolios had been better, yet our primary purpose is to examine the risk-on-risk-off decisions in isolation, and not how stocks performed compared to bonds.
Upon a Risk-On signal, we buy the risk asset. Upon a Risk-Off signal, we sell the risk asset and buy the safe asset instead. That is why Market timing signals are often referred to as Risk-On Risk-Off.
One prominent writer on market-timing strategies is quant investor Mebane Faber. Mr. Faber has written a paper titled A Quantitative Approach to Tactical Asset Allocation, in which he describes his test results of the following simple strategy proposed by Professor Jeremy Siegel: buy the S&P 500 when it is above the 10-month simple moving average, and sell the S&P 500 when it is below it. This way, you participate in the market when it’s trending up, but stay out of the market through its major downtrends. These occur when a major recession is taking place, but there are also quite a few false signals (also called false positives by academics).
The charts above show that not only Market timing seems to beat the S&P 500 from a returns perspective, but it also reduces volatility and lowers the maximum drawdowns by almost a half. The seemingly extraordinary result should be taken with a grain of salt, though. First, those results ignore the effect of transaction costs, taxes, and tracking errors. Realistically, the performance of such a market timing strategy would be equal or even lower than buy-and-hold the S&P 500’s. It is also important to note that the yearly reruns of the market timing strategy underperform the S&P 500’s yearly returns in about half of the years between 1901 and 2012. To be able to hold on to such a market timing strategy, you would need to have a deep conviction in it..and strong nerves.
Nevertheless, the ability to reduce volatility and drawdowns is such a compelling premise, that it triggered further research.
The semi-anonymous blogger Philosophical Economics (short research will reveal that his name is Jesse Livermore) has done a superb job dissecting Market Timing and risk-on risk-off signals. His articles (here, here and here) serve as a comprehensive introduction to this intriguing topic. Philosophical Economics discusses the academic efforts to find a signal which correlates perfectly with recessions, per their common academic definition. A theoretical perfect recession timing scheme is shown in the following figure (log scale):
As seen in the figure above, perfect recession timing strongly outperforms the market. It generates a total return of 12.9% per year, 170 bps higher than the market’s 11.2%. It experiences annualized volatility of 12.8%, 170 bps less than the market’s 14.5%. It suffers a maximum drawdown of -27.2%, roughly half of the market’s -51%.
Could we use fundamental macro indicators to detect, and therefore time recessions?
Philosophical Economics tries to devise a composite of several such indicators, which he refers to as GTT.
In his articles, he shows that the single indicator that yields the best results is the unemployment rate.
The following signal is thus proposed: Buy the risk asset when the unemployment rate crosses above its 12 months moving average, and exit the market when the unemployment rate process below its 12 months moving average. Let’s examine the results.
The Blue Line represents the S&P 500 timed with the unemployment signal. The green line represents the simple-moving-average strategy that we discussed initially. The unemployment rate signal seems to perform better than the simple-moving-average signal strategy. But can we improve it further? Can we devise a signal that uses both the unemployment rate indicator and the moving average of price data, to yield in even better results?
Well, it seems that we can. Paul Novell of Investing For A Living, A quant investing blogger and newsletter service provider, had tried such an approach before us, calling such new signal SPY-UI. The signal works in the following way. We sell the risk asset and buy the safe asset only if the unemployment rate exceeds its 12 months moving average, and the risk asset price goes below its 10-months moving average. This stricter rule is expected to reduce the number of false positives. On the opposite side, we buy the risk asset and sell the safe asset if one of the conditions reverse – Either the risk asset price goes above its 10-month simple moving average, or the unemployment rate falls below its 12-month moving average. We want to have less strict rules for re-entering the risk asset, so to enjoy the pullbacks which follow recessions, thus avoid being out of the market when it recovers.
Now that we have three promising risk-on risk-off signals, SPY’s 10-month SMA, The Unemployment Rate indicator (UI), and the combination of the two (SPY-UI), let’s see how those three signals affect the strategies we have on the website.
The timeframe for our testing is June 30th, 1999, to June 30th, 2018, a period of 19 years. This specific time frame was selected so it will be consistent with the other backtesting results on our website, thus allow meaningful comparison between strategies. We also conform to the practice in academia of using June 30th as a starting point for backtests.
Before utilizing market timing on quant strategies, let’s establish a baseline by testing our signals on the S&P 500 ETF – SPY. Here are our results:
The chart above shows that the S&P 500 gained 5.67% per year during our 19-year period. Its volatility, measured by the standard deviation of monthly returns, came in 15.55%, and the max drawdown -54.6%. Implementing the SMA 200-day market timing rules reduced the performance to 5.39%. This result is not uncommon. Similar to Philosophical Economics’ and Mebane Faber’s discoveries, we also discovered that market timing did not ALWAYS improve returns. Sometimes, market timing results in higher returns, and sometimes, it results in lower returns. As for volatility and drawdowns, the story is different. Here we can see a clear advantage to market timing, reducing the standard deviation to 9.13% (vs. 15.55%) and cutting down the max drawdown by about a half to a mere 23.5%. Not surprisingly, the Sharpe ratio has increased to 0.62 versus 0.43 for the buy-and-hold case.
Using the unemployment rate as a market timing signal yielded surprisingly good results. The average annual returns came in 9.04%, more than 3% higher than the buy-and-hold strategy. The volatility was significantly lower than buy-and-hold, albeit not as good as the SMA 200 case. Drawdowns are the smallest of the four experiments, at 18.6%. Using the SPY-UI signal did not result in a materially different outcome. Its returns are the highest, at 9.96%, yet its volatility is also slightly higher than with UI, at 11%. Its Sharpe ratio is the same as the UI case, at 0.92.
The Signal Count is an important metric. It indicates how many times we needed to enter and exit the risk asset in order to implement the market timing. For the SMA 200 case, we needed to buy or sell S&P 500 63 times over the period of 19 years, for an average of 3.3 times per year. Implementing the UI and SPY-UI signals would have been easier, with only 13 and 17 entries and exits during that period, less than once per year, on average.
A look at the chart shows that market timing signals have helped us avoid being invested during the two recessions of 2000-2003 and 2008-2009. That is desirable. But most of the other entries and exits had occurred in less favorable times, being false positives, or in layman language – false alarms. A practitioner of market timing has to take into account that false alarms are unavoidable and that they will result in unnecessarily lowering returns during normal times. Bleeding money on false alarms can be regarded as an unavoidable tax to pay for the chance of desirably being out of the market on major recessions and pullbacks.
Let us now examine how our quant strategies performed using market timing signals.
We begin with the Quantitative Value strategy. We use a portfolio consisting of 30 stocks with a market cap in the 40% (approx $200M )or higher, rebalanced yearly:
And on a log scale:
Contrary to the S&P 500 case we presented earlier, using the SPY-UI signal for QV did not impact the average annual returns in any meaningful way. It came in slightly lower, at 18.63% vs. 18.93% for the buy-and-hold case. The volatility is lower in all three market timing implementations compared to the buy-and-hold case, with the SPY-UI having the highest of the three. Sharpe is very high for the SPY-UI case, at 1.45. The results indicate that using SPY-UI to time the Quantitative Value portfolio may be desirable. But there is an important caveat, revealed when examining the price charts above. The SPY-UI (yellow) line opens a significant negative gap vs. the buy-and-hold case, between 2003 and 2007. A gap that continues to widen for years, and closes only when the 2008 recession strikes. A market timer must acknowledge that she will underperform buy-and-hold investors for prolonged periods of time, occasionally by a significant (negative) margin. The tradeoff then becomes – reduce volatility and drawdowns, while relinquishing some of the return potentials. Given that QV widely outperforms the benchmark, my personal opinion is that giving up a small portion of the return potential in exchange for gaining peace of mind and drawdown protection – is an easy choice.
The results for a Deep Value portfolio (based on VC2 ranking system), 30 stocks, 60% market cap (~$200M and up), rebalance yearly, is provided below:
And on a logarithmic scale:
The figures above show a trend consistent with previous results. Using market timing, the returns are lower, but so are the volatility and drawdowns. For this quant portfolio, the SPY-UI result in a higher drawdown than the simple UI case. Similarly to QV, the Sharpe ratio is the highest for the SPY-UI, at 1.07.
The market timing signals perform even better for a low EV/EBIT portfolio:
In contrast to the VC2 portfolio, the average returns of the EV/EBIT portfolio with SPY-UI signal are better than the baseline strategy. Coupled with the max drawdown, which was cut in half, and with the lower volatility – the Sharpe ratio of 1.14 is about 25% better than the 0.91x ratio of the buy-and-hold strategy.
Let’s now examine the Quantitative Momentum strategy:
On a log scale:
The returns of the SPY-UI portfolio are roughly the same as for the baseline strategy, at 16.6%-16.7%. Max drawdown was significantly trimmed down by almost two thirds, and the volatility is sufficiently lower. Indeed, market timing takes away some of the headaches that are involved with momentum investing.
Lastly, let’s see the effect of market timing on the most conservative strategy we track, The Graham’s “Defensive Investor” strategy. The results are surprising:
With a log scale:
The SPY-UI did improve the average annual returns, but only by a small amount, 10.74% vs. 10.02% for the buy-and-hold strategy. The max drawdown did not change by a significant amount. Volatility is lower for the market timing portfolios, but not by a wide margin as with the previous strategies.
One could recognize a pattern here. The more volatile and “risky” the baseline strategy, the stronger the smoothing effect of using market timing. That would be an anecdotal claim, lacking a strong statistical backing. It may be true, but we have not collected enough evidence to make such a claim confidently.
What can we conclude from the examples discussed above? Is using market timing the desired practice to protect portfolios from large downtrends?
The examples above show that using market timing, and especially the SPY-UI signals, contributes to lowering the volatility and max drawdowns of portfolios. As for the absolute average annual returns, in some cases, they are improved (Graham, EV/EBIT, S&P 500), and in other cases, they are impaired. However, in all the examples we examined, the risk-adjusted returns, as measured by the Sharpe ratio, show improvement over the buy-and-hold strategy.
Market Timing adds a maintenance effort on the portfolio owner, requiring him or her to buy and sell holdings whenever the signal triggers. Most of those signals are false positives (i.e., false alarms), but the true alarms have kept investors out during the major two recessions that occurred during the last 20 years. Note that every such signal results in transaction costs, and every sale may result in a tax liability (which is deferred in the buy-and-hold case).
By now, you may be thinking that market timing’s advantages clearly surpass buy-and-hold and that it is a free lunch that will enable you to sit out of major recessions.
Well, not so fast. There are a few caveats to be discussed.
First, the results of the experiments above are not statistically significant. They serve as evidence (that market timing works), but not as proof. During the last 20 years, we have suffered only two major recessions. Market timing signals were triggered between 13 and 63 times during those years. That is not a large enough sample base to be statistically meaningful. The true positives are far and in between, and our backtesting period is too short to cover all or most of the potential scenarios. Philosophical Economics backtests, ranging over 90 years, are better, but even they are too short.
Secondly, we run a risk of data-mining (a.k.a. curve-fitting). While we had found the signals to work extremely well on past data, their applicability to future data is questionable. State differently, we are looking at the rearview mirror and solving the problems of the past, and not those of the future. A powerful way to reduce the risk of data mining is to run the strategy out-of-sample data, such as on different benchmarks, portfolios, and even individual stocks. Philosophical economics has done testing on individual stocks, and his results were mixed. Some stocks benefited from market timing, while others – clearly did not.
Lastly, there is the psychological effect. An investor would have to be mentally prepared market timing for extended periods of time, only to discover that the entry and exit signals were false alarms, derailing performance. Think of yourself going through a 10-15 years period of executing false signals resulting in transaction costs, taxes, and (sometimes) lower returns than buy-and-hold.
Yet the gains could be substantial if eventually, a recession occurs, and market timing keeps an investor from seeing his or her capital lose half or more than its worth during a major recession.
Naturally, we cannot wait 1000 years to gain more statistics and out-of-sample data. We have to make a decision now whether to employ market timing or not. A lack of decision is a decision by itself (against market timing). Allow me to propose how to think about it.
If you seek to minimize the time you spend on managing your investments, assuming you have a long investing time horizon – feel comfortable to avoid market timing altogether. Especially if you are the type of investor that does not check stock prices hourly, or daily.
Alternatively, if market pullbacks scare you, making you unsure that you can wether seeing your capital go down abruptly (even if only temporarily), consider market timing. Better use market timing intelligently using mechanical signals than timing the market by being triggered by fear and grid – bailing out after major declines, and re-entering after major rebounds.
If you are not sure which type of investor you are, consider splitting the odds. Upon a risk-off signal trigger, sell half of your portfolio positions value. Upon a risk-on signal, return to be fully invested.
The choice is yours. There is no one solution best for all.
During the next few weeks, market timing signals will be added to the Premium membership program. Stay tuned.
Market Timing Resources
A helpful introductory resource is the Wikipedia entry on Market Timing. It reminds us that market timing is a controversial topic and a practice that was NOT proven beyond a reasonable doubt to be a successful endeavor. On the topic of over-optimization, Wikipedia writes:
“A major stumbling block for many market timers is a phenomenon called “curve fitting“, which states that a given set of trading rules tends to be over-optimized to fit the particular dataset for which it has been back-tested. Unfortunately, if the trading rules are over-optimized they often fail to work on future data. Market timers attempt to avoid these problems by looking for clusters of parameter values that work well or by using out-of-sample data, which ostensibly allows the market timer to see how the system works on unforeseen data. Critics, however, argue that once the strategy has been revised to reflect such data it is no longer “out-of-sample”.
A good introduction, though a bit lengthy, to the topic of market timing is the academic paper by Mebane Faber, A Quantitative Approach to Tactical Asset Allocation. The paper provides detailed results on the S&P 500 performance with and without the use of the simplest market timing signal – using a long simple moving average to enter and exit the market. Faber’s finding matches ours. Using simple moving average works well on the S&P 500 over a very long-term, but underperforms it for half of the 1-year periods. Here is one of the important results in the paper:
Examine the yearly returns, especially since 2006 (the out-of-sample period of the research). While market timing avoids the big drops in market prices, it achieves a lower return in at least half of the years.
Another interesting introductory article to the subject is one written by David Ott and published in Alpha Architect blog, The Dirtiest Word In Finance: Market Timing. Ott’s thesis is simple. Managed futures funds have a track record of beating the market. Managed futures funds use market timing. Conclusion – market timing should beat the market.
Now that we covered the basics, we are ready to go even deeper. Philosophical Economics blog (which we believe is authored by Jesse Livermore), has done the most comprehensive work we have found on Market Timing. His set of articles are nothing short of being superb. We applaud the depth of his analysis.
In his first piece, Trend Following In Financial Markets: A Comprehensive Backtest, he describes the backtesting results of roughly 235 different equity, fixed income, currency, and commodity indices, and roughly 120 different individual large company stocks (e.g., Apple, Berkshire Hathaway, Exxon Mobil, Procter and Gamble, and so on). The results show that market timing works well on indices, but works poorly on individual securities. Philosophical Economics also includes the best description we have found on the requirements which should be imposed on a quant strategy for it to be dependable. His eye-opening description is not related only to market timing strategies, but to any mechanical strategy. He claims that any strategy should meet the tests of being Generic, Efficient, Long-biased, and Recently-successful. We have taken a note to write an article on those four requirements alone. How does the moving-average based market timing strategy fare against those four requirements? Well, it is generic, long-biased, and recently successful, having outperformed since the 1960s. But its efficiency is questionable. It switches on 10% of the months, and 75% of those switches are false positives, contributing nothing (and even negatively) to the end results.
In his second piece, In Search of the Perfect Recession Indicator, Philo Econ explores economic indicators that could be used as risk-on risk-off signals. He recaps on the use of a composite of 6 economic indicators (aka GTT) aimed to detect a recession, and also discussed using the single indicator of the unemployment rate. GTT turns out to be a lagging indicator, signaling to exit the market only halfway through the recession. On the contrary, the single indicator of the unemployment rate is a good proxy for a recession, having an 86% correlation to recessions.
As we mentioned above in the Spotlight section, Paul Novell of Investingforaliving.us blog has also done some research on market timing models, including GTT and SPY-UI. His first step is to summarize Livermore’s results and replicate them on his own. Then, he tries out SPY-UI with several bond ETFs as his safe asset. Finally, he presets SPY-UI’s results with his quant strategies. Novell’s results resemble ours in many respects, which is not surprising.
The resources above present examples of market timing working well. But, as we’ve indicated above, market timing is not easy, not proven scientifically, and does not always work well. We cannot end this section without providing two resources which support our claims. The first, a Forbes piece titled Busting The Myth Of Market Timing is an easy-read but not too deep. The second piece, Ben Carlson’s post, Market Timing is Hard, is a deeper and an excellent perspective on the matter.
Reading the resources above, along with our Spotlight article, will provide members with enough insight to develop an opinion on market timing and decide on the extent of its application in his or her portfolios.