Can Timing Recessions Improve Asset Allocation? A Back Test Report

by: Jason Cawley


A previous article gave a recession predictor model and explained its construction.

Here I give a method of using of that predictor to guide asset allocations for accumulation and distribution phase investors.

Back tests of the method show large gains to expected return with major reductions in risk for both categories of investor.

As with all back tests, caution must be used in interpreting the results, but the method is promising.

In a previous article, I gave a simple statistical model designed to predict US real GDP recessions a year ahead of time, using a few simple macroeconomic times series as predictors. You can find that previous article here.

Here I want to show how one might use such a model to guide asset allocation decisions. This is frankly an exercise in market timing, which most academics tell us is impossible - and I will return to the potential pitfalls in the whole method at the end of this article.

The motivation, though, is to use the fact that US recessions don't appear out of the blue but are preceded by a predictable pattern of macroeconomic indicators to give us advanced warning of at least the most important stock market decline periods. We then shift our asset allocations to a more defensive or "risk off" stance when our recession warning indicator "flashes", and go back to a fully invested "risk on" stance when that indicator has given us the "all clear" to do so.

I present below the detailed operational rules I recommend for the specific recession predictor function given in the previous article. Other such predictors exist and the approach could be adapted to using them instead, but that is beyond the scope of this article. My intention is also to give sensible overall asset allocation approaches to two distinct kinds of investor, the accumulation phase and the distribution phase respectively, and to show how they should use such a warning system to shift their asset allocations in a counter cyclical manner.

Asset Allocation Rule Cumulative Returns

Then I report the results of back testing the proposed rules on the previous 20 years of US market history, which are previewed in the chart above. This involves the warning system telling us to reduce stock allocations before the two previous recessions, the one in 2001 and the one in 2008. Since that model was trained on the last 40 years and the last 5 or so recessions, it unsurprisingly does a good job warning of the approach of the last 2, but the details of the warning timing, stock market behavior near recessions, and the like are not trivial. I report the specific periods in which the rule proposed leads us to a "risk off" position for those two previous cases and discuss the event timings seen in each.

Then as the meat of the backtest, I will show the cumulative investment performance of following the trading rule for accumulation or distribution phase investors, which I compare to all stock or constant 60-40 stock and intermediate corporate bond asset allocations over the same period. I give a table of the average annual rate of return and the risk (annualized standard deviation of monthly returns) for each approach, as well. I discuss the lessons of the results, as well as some of the ways they might not generalize to future performance.

To begin with then, by an accumulation phase investor I mean one still adding to portfolio savings at a decent rate, and expecting to continue doing so for at least 10 years before calling on any income from that portfolio to finance consumption. By a distribution phase investor I mean one using investment income to fund consumption, especially a retired investor. A distribution phase investor has a greater need for stable income and less ability to support temporary drawdowns, whereas an accumulation phase investor can afford to focus on expected return more than risk, to a greater degree. Both should still care about risk, of course.

For either investor, I recommend that 20% of the portfolio allocation always stay in stock regardless, and similarly that 20% of the portfolio always stay in intermediate corporate bonds, regardless. These may be combined in a core balanced mutual fund position, if desired. Keeping some position in those asset classes at all times means that normal rebalancing will tend to increase weight of asset classes that have fallen in relative price and sell some of those that have outperformed recently. In addition, for the distribution phase investor I recommend a constant weight of 20% of the portfolio in long term income investment, whether a corporate bond fund, preferred and income fund, or an appropriate low cost annuity. This gives income stability even if rates move significantly lower for a prolonged period of time.

This then means that an accumulation phase investor is only going to move at most 60% of their portfolio weight in response to timing signals of "risk on" vs "risk off", and a distribution phase investor only 40%. My recommendation is that 40% be either in stock in the risk on phase or in short term Treasuries in the risk off phase, as the result below use those instruments. One could in principle use other cash investments or CDs, or short term corporates in preference to Treasuries. 2 Year Treasuries, though, have the characteristic that they tend to do well when stocks are falling, while corporates can experience higher spreads at those times. And they have a higher average return in the past than cash proper.

For the remaining 20% of the accumulate phase investor, my recommendation is to have that also in stock in the risk on phase, and to have it in long term income investment in the risk off phase. The goal there is to get into such long income only near cycle highs for interest rates and as the yield curve inverts, while also dialing back stock market risk at those times. This is less defensive than moving that portion as well into 2 year Treasuries.

Summarizing these rules, an accumulation phase investor in a risk on period is 80% stock and 20% intermediate corporate bonds. In a risk off period he is 20% stock, 20% long income, 20% intermediate corporates, and 40% short term Treasuries. A distribution phase investor in a risk on period is 60% stock, 20% long income, and 20% intermediate corporate bonds. In a risk off period that moves to 20% stock, 20% long income, 20% intermediate corporates, and 40% short term Treasuries - the same allocation in those periods as the accumulation phase investor.

So what defines a risk off period, in terms of our previous recession warning rule? The answer is that a risk off period is initiated whenever the recession warning indicator exceeds 0.40. Once such a signal has been received, it remains in effect until 1 full year after the recession warning indicator has fallen below a score of 0.20. The year delay is very important - our warning indicator was designed to give at least that much warning, and it can and it will fall to low levels again once short rates fall on the onset of crisis conditions. That is not a signal to jump back in - on the contrary. A full year after the "all clear" score below 0.20 is the first time one can safely get back into stocks. If one likes, one could also wait to see that the recession is over, but the tests below apply the 1 year after 0.20 risk score rule.

What are the results of that rule for the last 2 recessions? The answer is that the method first sends the risk off signal in November of 1999, and then leaves it in the risk off position until July of 2002 when buying back into stocks is indicated. The next warning is set in June of 2006, and remains on until January of 2009. All the rest of the time in the past 20 years, the system sent a risk on, OK to invest, signal. It is still sending that message at the present time, though the yield curve indicator in particular now bears watching, as discussed in the previous article.

Notice, these times are not set by perfect times to get in or out of stocks, but by the times our recession warning indicator reaches its threshold levels, with the year lag for getting back in. The actual market top for SPY (our stock index used here) was September 2000 and its bottom wasn't until March of 2003. In the later recession, the top was in October of 2007 and the bottom not until March of 2009. In both cases the warning rule gets us out early and gets us back in a few months before the final bottom.

I am deliberately not adjusting the rule given for these details, because one cannot expect them to hold perfectly in future cases. That would be "overfitting" the back test to the precise timing details of just a pair of previous market cycles and the specific timing of those relative to both the recessions themselves and the warning system's lead. In the first of the two, the stock market peak came 3 months after the warning and 9 months before the actual recession. In the second, the stock market peak came 16 months after the warning and right at the start of the actual recession - which was not actually recognized as starting then until a quarter later. We cannot expect the time alignment of these things to always follow the same specific spacing - that is why we targeted a full year's warning and don't adjust the method to such specific case details.

Here is the table of actual average returns and risks this rule yields. For comparison I give the average returns and risks of 100% stock (NYSEARCA:SPY), a constant 60-40 balanced allocation between SPY and intermediate corporates bonds, and 100% such intermediate corporate bonds. The last 2 columns give the relative return and risk compared to the all stock figures from the first time, expressed a ratios.

Risk Reward Table

The table already shows the sense in the conventional approach of a 60-40 stock bond allocations, compared to all stock. That gives a large reduction in investment risk for a trivial reduction in expected return over this period. As the following line shows, however, this depends in part on the fine return that intermediate corporate bonds gave over this particular 20 year period - nearly 6% annually. That reflects higher interest rates at the start of the period and a large move lower in those interest rates. We cannot expect the same to recur over the next 10 years, since we are no longer starting from a 6% rate level, but only half that. It is worth understanding, however, that over the past 20 years intermediate corporates "shot the lights out" in risk-reward terms, and a modestly leveraged position in them would have handily beat stocks with less overall risk, over that past period.

The last two lines show the results of our actual back test. The accumulation phase version of the rule beats all stock by 3% a year, while having only about the risk of the 60/40 balanced portfolio. The distribution phase version achieves 1.8% per year lower risk along with 2.37% per year greater return than the 60/40 balanced portfolio. Recall that the distribution phase rule is only 60/40 in the risk on periods; one would expect a significant reduction in risk relative to that benchmark, but the better expected return as well is a clear superiority of successful defensive timing. Provided, of course, that the timing is succesful.

Can we be sure that these back test results would be seen in future uses of the recommended method? No, for two different reasons. One, as mentioned the return of the bond components that we shift to, out of stocks, likely will not be as good in future periods as they were in our back test. All allocation systems that put on higher bond weight will look better in a risk reward since when tested over any period when bonds had about the same return as stock for lower risk. Two, our recession spotting system was trained in part on the past two recessions, so of course it can spot them coming successfully and get us out of stock at the proper time. We do not know that the next recession will have exactly the same pattern, or that any associated stock market decline will occur with the same broad timing relative to the next recession that we saw in the last two.

Nevertheless, I think the back test shows considerable promise to such methods. Stock market returns routinely exceed bond returns for sound reasons likely to recur. Those higher long run returns are a payment to incurring significantly higher risk than bond investment involves. Those risks are in turn highly concentrated in the time domain. While every correction is not associated with large macroeconomic events like full blown real GDP recessions, the biggest and most damaging ones are. And those events are big enough and take enough time to build that it is quite plausible they can be spotted a reasonable distance off.

I close with some warnings about using such systems. They require a fine mix of dispassionate objectivity with a skeptical understanding of their limitations. It is very easy to be swayed by sentiment or other indicators, lose faith in what the warning system is saying, and get out too early, or worse back in too soon thinking that the system has sent a false signal. It is clearly possible for the next recession and any associated market declines to deviate from the past pattern the system expects. But if one second guesses every warning the system sends, one ends up trading on emotion instead of what the data is actually saying. The system above is constructed with these limitations in mind - that is why it modestly varies only about half the portfolio is response to warning signals, is constructed to give broad warnings far enough ahead of time, etc. Your mileage will certainly vary, and please use with due caution.

I hope this is helpful.

Disclosure: I/we have no positions in any stocks mentioned, and no plans to initiate any positions within the next 72 hours.

I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.

Additional disclosure: Long US stocks, preferred, and intermediate corporate bonds.