Market Timing Strategies

Add to del.icio.us

Introduction

In August of 2008, the S&P 500 index was valued at about 1,300 – seven months later its value had plummeted to less than 700. Many investors who followed a strategy similar to Sigmaís recommended strategy would have remained invested in equities before, during and after this crash. Those investors would have benefited from their systematic rebalancing by pulling out some profits at the top and buying shares at low prices at bottom. However, this portion of their portfolio would have undoubtedly suffered a serious setback. Could this have been avoided? We knew in the summer of 2008 that the long term PE ratio (calculated using average earning over 10 years) was hovering above 26 Ė thatís more than 1 standard deviation above its long term average. This ratio reached close to 45 in 2000 right before the collapse of the market. Didnít that clearly indicate that equities were overvalued at these times? Shouldnít investors have reduced their allocation to domestic equities based on this type of information?

In this article, we will show that market timing turns out to be very difficult to effectively implement. We will spend some time examining one reasonable strategy that has been shown to have some effectiveness in past research. However, we will show that the benefits of this approach are very modest and would likely lead to lower returns. Although it’s tempting to believe that you should do better by periodically moving assets out of overvalued asset classes, we think that you are best served by not using market timing strategies.

Table 1

Chart 1: PE Ratio with earnings calculated as 10 year historical average.

In this article, we will take a quick look at the potential to use a more dynamic asset allocation strategy which will shift money out of asset classes which appear overvalued and move money into undervalued classes. This is a classic market timing approach. It is based on the belief that we can determine when the market will probably rise or fall in the intermediate term. If weíre correct, we should be able to profit handsomely versus a strategy focused on holding and rebalancing periodically.

Evaluating market timing-based strategies is a big question. Most of the research has indicated that it is very unlikely that an investment manager can successfully time the market (see Charles Ellis 1985). However, given the marketís recent collapse and our interest in the predictive power of the long term PE ratio, weíll explore a couple of strategies that seem very reasonable to us.

Hints of Evidence

Before we dive into some analysis of some proposed strategies, letís discuss a couple pieces of data that offer hints that successful market timing might be possible. First, letís take a look at the expected future returns of the S&P 500 when we condition on the value of the long term PE ratio. Table 2 shows the actual average 1 month, 1 year, 5 year, 10 year and 20 year returns for almost 70 years of data when we bucket the starting point into deciles based on the initial long term PE ratio. Youíll note that the future returns are well ordered based on decile Ė so, the periods with the lowest initial PE ratios end up having the highest future returns. This piece of data seems promising. If we know the initial long term PE ratio (which we do), then we should have a pretty good sense of what type of medium and long term returns to expect. Thus, if the medium term returns donít look great, maybe we should shift some money away from this asset class. While this table of data seems fairly logical, it could have profound implications on our ability to time the market.

PE Decile Average 10 yr PE 1 Month Return 1 Year Return 5 Year Return 10 Year Return 20 Year Return
1 9 2.1% 28% 18% 16% 15%
2 10 0.8% 19% 16% 15% 14%
3 12 0.9% 18% 14% 15% 14%
4 13 0.9% 13% 14% 14% 13%
5 15 0.8% 9% 11% 13% 12%
6 17 0.5% 7% 8% 11% 11%
7 19 0.5% 4% 8% 10% 11%
8 21 0.2% 5% 13% 9% 9%
9 24 0.7% 7% 6% 6% 9%
10 34 0.7% 4% 3% 5% 8%

Table 2: Average annual returns of the S&P index based on initial PE ratio. (Note: The one month return is not annualized)

If domestic equity returns in the future donít look promising, then one place to shift your allocation would be into treasuries. The next logical question would be whether we might be able to predict medium term treasury returns in a similar fashion as we did for the equities. Since we donít have long term PE ratios for treasuries, we could instead use the current rate of return to bucket them into deciles. Take a look at Table 3. Here we see some logical results Ė treasuries with the largest current return will have the largest expected medium and long term returns. Thus, again, it seems like we can have a pretty good understanding about our future returns by simply knowing some readily available current information.

Treasury Decile Average T-Bill Yield 1 Month Return 1 Year Return 5 Year Return 10 Year Return 20 Year Return
1 0.1% 0.0% 0% 0% 0% 1%
2 0.4% 0.0% 0% 0% 0% 1%
3 1.1% 0.1% 1% 1% 1% 2%
4 1.9% 0.2% 2% 2% 2% 3%
5 2.9% 0.2% 3% 3% 4% 4%
6 3.8% 0.3% 4% 5% 5% 6%
7 4.8% 0.4% 5% 6% 6% 7%
8 5.5% 0.4% 5% 6% 6% 6%
9 6.9% 0.6% 7% 7% 7% 6%
10 10.1% 0.8% 10% 8% 7% 6%

Table 3: Average annual returns of the T-bills based on initial yield. (Note: one month return is monthly)

While both these pieces of information seem to indicate that there may be a way to construct a market timing-based asset allocation strategy that would significantly outperform a more passive maintain and rebalance strategy. In the next section, we will examine some intuitive, reasonable strategies and see how they fare in practice.

Results of Market Timing Trading Strategies

Now, we will focus on an interesting market timing strategy that relies on the above information to give logical advice to the investor about when to move into or out of the domestic equity market.

Standard Yield Differential

The Standard Yield Differential is an interesting statistic upon which one can create a strategy for dynamically determining portfolio allocations based on an intuitive combination of long term PE ratio values and treasury yields. This model was first discussed in the academic literature by Wong et al. (2001). They define the standard yield differential (SYD) as the value of the long term EP ratio (the reciprocal of the PE ratio) minus the current treasury yield. Letís consider what various levels of the SYD might indicate. If the SYD is high, this implies that the PE ratio is low (thus the EP is high) and the treasury yield is low. Thus, it logically seems like a good indicator to move into equities. On the other hand, if the SYD is low, this means that the PE ratio is high (EP is low) and the treasury is high. This would indicate a good time to remain in treasuries.

SYD = 1/PE – Tyield

We like this simple model for several reasons:

  • The long term PE ratios appear to have good medium term predictive power for equity returns
  • The current treasury yield appears to have good predictive power for the medium term treasury returns
  • The SYD is a very intuitive and logical allocation strategy
  • The SYD can be very easily calculated and, therefore, an allocation based on this figure would be relatively easy to implement

It appears like this might be the type of quantitative model that could produce a market timing-based dynamic asset allocation model that would outperform a fixed allocation strategy. Next, weíll look at how one implementation performs against historical data.

Academic Studies

In 2002, Shen released a study using the SYD to determine when a portfolio should be invested in equities and when its allocation should be moved to treasuries. His method was to move his portfolio from equities to treasuries whenever the value of the SYD was below its historical 10th percentile value. This decision rule ensures that the investor will be invested in equities for the vast majority of time periods. The allocation decision was made on a monthly basis. In other words, he is trying to keep money in equities except when a very low historical value of the SYD occurs. In his analysis, he calculated two different versions of the SYD Ė one based on the yield of a 3 month treasury bill (the short model) and one based on the yield of the 10 year treasury bond (the long model). He looked at 30 years of data from 1970 to 2000. See his results in Table 4. As you can observe, his strategy based on the SYD short model achieved an average monthly return of 1.5% while remaining invested in equities for 78%. This average return significantly beat the return of remaining invested in equities for the entire time period. Additionally, it produces a better Sharpe ratio. Thus, Shenís results suggest that an asset allocation strategy based on the SYD can produce significantly better returns than merely remaining invested in equities through thick and thin.

Strategy Time Period Monthly Return Sharpe Ratio Time in Equities
All Equity Jan 1970 to Dec 2000 1.1% 0.13 100%
Shen – SYD Short Jan 1970 to Dec 2000 1.5% 0.21 78%
Shen – SYD Long Jan 1970 to Dec 2000 1.3% 0.16 76%

Table 4: Returns of SYD strategies reported by Shen (2002).

In 2005, Brooks et al. wrote a paper where they examined the potential of several strategies which attempt to time the market. They included the two Shen SYD models in their analysis. They used a much longer time horizon to evaluate their models (from 1927 to 2003) than Shen. Their results can be found in Table 5. You can see that the SYD short model again outperforms the fully equity invested results but the difference is very small. The SYD long model actually underperforms the strategy of remaining in equities during all time periods. This casts doubt on the value of using an SYD-based strategy. It does improve returns and Sharpe ratio slightly but you must choose the correct version of the SYD model. An obvious question to ask would be: why do the results of the SYD model differ significantly when the time period is extended? We believe that the problem lies in Shenís analysis as we will explore in the next section.

\
Strategy Time Period Monthly Return Sharpe Ratio Time in Equities
All Equity Jan 1927 to Aug 2003 0.94% 0.13 100%
Brooks et al. – SYD Short Jan 1927 to Aug 2003 0.96% 0.14 88%
Brooks et al. – SYD Long Jan 1927 to Aug 2003 0.88% 0.13 82%

Table 5: Returns of SYD strategies reported by Brooks et al. (2005).

Sigma’s Results

Since these two academic papers produced such different results, we thought it prudent to conduct our own analysis to validate their results. Using our own equities and bond data from Shiller, we implemented the various models and our results can be found in Table 6. The first point to note is that our results for the SYD short model with Shen’s time periods were much different than what he reported. In fact, they are almost equivalent in average return to a strategy of remaining in equities at all periods (although the SYD short does produce a better Sharpe ratio). Our figures matched the Brooks et al. analysis very closely. After checking and re-checking, we are convinced that Shen made an error in his analysis. Thus, it appears that implementing allocation strategies based on the SYD wonít produce better overall returns but they might reduce volatility and increase the Sharpe ratio.

\
Strategy Time Period Monthly Return Sharpe Ratio Time in Equities
All Equity Jan 1970 to Dec 2000 1.1% 0.16 100%
SYD Short Jan 1970 to Dec 2000 1.1% 0.17 79%
All Equity Jan 1927 to Aug 2003 0.94% 0.13 100%
SYD Short Jan 1927 to Aug 2003 0.92% 0.13 88%
SYD Long Jan 1927 to Aug 2003 0.76% 0.11 70%

Table 6: Sigma’s calculated returns of SYD strategies.

Explaining the Performance of the SYD Short Strategy

Now, we will examine why these returns for the SYD-based approaches donít live up their promise as a means of identifying future periods of high or low equity returns and allocating your portfolio accordingly. First, letís examine when Shenís SYD short model indicated that the investor should have moved his funds into treasuries. In Chart 7, the green boxes represent periods when the portfolio would have been invested in treasuries.

Table 1

Chart 7: Time periods invested in T-Bills when following the SYD short strategy.

When we examine the data, the moves into treasuries worked out well in 1983 and 1990. These were periods when treasury yields were quite high so substantial returns could be gained in treasuries while equities suffered as predicted. However, the moves into treasuries in 1997 would have missed out on the strong gains in equities from 1997 to early 2000. This might be tolerated by the investor since the market suffered a steep decline in later 2000. The investor would have been shielded from this decline. However, this occurrence demonstrates an issue with attempts at market timing. While the investor would have missed the steep decline in mid-2000, they also missed out on the strong equity run-up of the previous three years. These models canít predict when the top of the market will occur Ė they merely indicate when the opportunity for returns might be better in treasuries versus equities. Often, the equity market will continue to rise after this indication and then will decline in value until equities appear more attractive. So, the model will cause investors to miss some of the climbs to the top of the market while they are invested in lower performing treasuries. Now, it is worth remembering that the value of equities is measured not only by price appreciation but also by dividends earned when holding the equities. Thus, it turns out that missing out on the dividend yields from the equities was the reason why holding treasuries during this period (1997-2000) didnít result in a greater return. Treasury yields were quite low during this period. Thus, it seems that the SYD works fairly well when the absolute treasury yields are on the high side. Otherwise, the potential to miss out on dividend payments while sitting out the ride up and down of equities might end up being more painful to the investor.

So, we have found some fundamental problems with this model. First, it will cause investors to miss out on some potentially valuable equity run ups. That might be acceptable if the returns of the treasuries would more than compensate for lost dividend payments. But, this model doesnít consider the rate of dividend payments or the absolute level of treasury yield. Perhaps some improvements on the model (increasing its complexity) would produce more positive results? Perhaps not?

Summary of Issues

While we have focused on only one core market timing strategy, we do think that it has pointed out some of the pitfalls and challenges of market timing-based investment strategies. In fact, the SYD strategy is one of the few strategies which has some evidence pointing out that it may work (although we do doubt the validity of Shen’s analysis). There are a few caveats that we should mention about our analysis. First, since, the average annual return of treasuries is substantially lower than domestic equities (represented by the S&P 500), it may be difficult to construct an asset allocation strategy which uses the safety of treasuries to increase returns. These strategies seem to effectively increase Sharpe ratios but gaining an edge in returns may prove very, very difficult. However, if we explored moving funds into an asset class (like foreign equities) that have return characteristics closer to domestic equities then we might find some success. However, it is likely that a model shifting allocations between two asset classes with similar return characteristics would produce its fair share of issues that might hamper returns versus a strategy of fixed allocation with periodic rebalancing.

As Charles Ellis points out “the evidence on investment managers’ success with market timing is impressive – and overwhelmingly negative”. Most logical market timing ideas actually perform worse than our recommended strategy of picking an asset allocation and maintaining it. Without doubt, we think that there are some more sophisticated models which can probably slightly increase your long term returns. However, we don’t recommend attempting to time the market or apply tactical asset allocation. It is difficult to find a model which will perform well in future periods (versus simply testing well on historical data).

Your best bet is to determine some fixed allocation percentages among asset classes and to rebalance periodically.

References

Chris Brooks, Apostolos Katsaris, and Gita Persand. “Timing is Everything: A Comparison and Evaluation of Market Timing Strategies”. 2005
Charles Ellis. Investment Policy. 1985
Pu Shen. “Market Timing Strategies That Worked”. 2002.
Wong, W.K., Vhew, B-K., and Sikorski, D. “Can the Forecasts Generated from E/P Ratio and Bond Yield be used to Beat Stock Markets?”. 2001