In a recent Seeking Alpha article discussing what I learned from the 2008 market debacle, I mentioned that I was working in the area of “regime switching.” As applied in this context, this refers to an approach that switches back and forth between bull- and bear-market strategies depending on market conditions. My interest in this approach stems from my belief that a major thing we learned from 2008 is that “market timing” needs to be elevated from its “nobody can really do that” image into a mainstream discipline worthy of serious attention, research, and application. Toward that end, I introduced what I refer to as a “Multi-market” model on Portfolio123.com.
This is a fairly complex protocol involving a two-part ranking system, a specially-created bear-market universe, and most buy-sell rules and rank factors being embedded within if-then statements. This is definitely not a penultimate expression of the regime switching approach. But I am sufficiently satisfied with the system’s major components and would now like to step beyond backtesting and simulation and into real-time evaluation. Besides introducing the model today, I'll update its progress periodically, so we can all see which aspects work and which ones require improvement.
This model consists of the following elements.
- A multi-market bull-bear ranking system
- A market-timing model that tells us which component of the ranking system should be used at particular points in time
- A specialized defensive-stock universe for use when conditions are bearish
- A super-bear timing rule that, when active, holds proceeds of stock sales in cash rather than reinvesting in equities
- A multi-market rank threshold that imposes a more stringent buy-hold threshold during bear markets
- A multi-market diversification rule that imposes different requirements for bull or bear markets
Further discussion of each element can be found in the Appendix below. But before examining that, I'll review simulation test results and look at what the model is telling us to do right now.
The simulation, conducted on Portfolio123.com, runs from 3/31/01 through 1/29/09. The portfolio is normally targeted to hold ten positions and is rebalanced weekly. I assume a per-trade $10.99 flat-fee commission as well as 0.5 percent price slippage.
Figures 1 and 2 show the simulated portfolio's returns over the course of the test period.
click to enlarge
There was some trivial underperformance in 2001 (the portfolio was all cash from inception till early 2002). Beyond that, there was one year of positive but lackluster performance relative to the S&P 500, 2006, when the model appears, in retrospect, to have been a bit too conservative for too much time. Otherwise, the returns look fine.
Figure 3 shows the most recent, and traumatic, 12-month period while Figure 4 provides a close-up of cash levels during this interval.
We saw from Figure 2 that the market fell by about 30 percentage points more than did the simulated portfolio. Figures 3 and 4 show us that a large portion of the model's relative strength came from being cash-heavy at the right times. But further relative strength occurred later in the year, when the portfolio was heavily or fully invested in stocks. Apparently, the bear-market portion of the ranking system and use of the smaller defensive-stock universe served us well.
This is important since market timing remains a challenging endeavor. The Portfolio123 timing model (click here for a Seeking Alpha discussion of it) is the basis for the one used here and has backtested well. But right now, the market is going through uncharted waters. So there are still questions for which we'll have no answers until more time passes and observations occur. A particularly poignant one for now involves the extent to which the initial stage of share-price recovery will occur before formal publication of higher earnings estimates.
Investors may react to economic developments before they get reflected in company guidance and analyst estimates. Given the timing model's use of five- and 21-week moving averages for evaluating consensus S&P 500 estimates, it is possible that it may stay bearish during the initial, and potentially vigorous, stage of a recovery, whenever that comes to pass. Of course, no timing model can be expected to be 100 percent perfect, so a once-per-cycle mis-step like that would seem quite tolerable. But from a human standpoint, that's much easier to say now, when recovery remains a future hope rather than a present reality, than might be the case when it's actually happening.
For this model, the key lies in the fact that it is not a simple all-cash or all-stocks approach. It is quite willing to hold equities, often up to being fully invested, even during bear markets. Instead of always running to cash, it simply switches to a more defensive set of equities. So it's encouraging to see that the bear-market version of the equity strategy did what it was supposed to do. That would also seem relevant for other occasions when less dramatic zigs and zags are missed. In other words, we're simply asking the timing model to serve as a traffic cop to direct the flow of logic among two different equity strategies. We're not asking it to carry the full weight of our entire investment program.
Figure 5 evaluates key risk measures.
|Note: The Standard Deviations, Sharpe Ratios and Sortino Ratios are based on daily price changes.|
The data shows that the above-market returns were accompanied by below-market risk.
Figure 6 sheds light on one more important piece of the puzzle, trading costs. This is always an issue, and is especially noteworthy with weekly rebalancing. The after-commission performance was fine. But if one were to actually implement a multi-market strategy like this, choosing a very-low-cost brokerage firm like FolioFN or even one of the lower cost traditional on-line brokers could have a significant impact on results.
We could also cut commissions by switching to a four-week re-balancing period. For now, though, I'm reserving judgment on this approach. Trading costs aside, one-week rebalancing is comforting in that it allows the portfolio to unravel bad investments more quickly. But there's the flip side: it provides more frequent opportunities to erroneously unravel good trades. The latter has shown itself to be an issue in various tests I conducted.
What to do now
The last time the model bought stocks was 11/24/08. Table 1 lists the stocks that entered the portfolio at that time. Notice the industries; these are all from the defensive universe, meaning that the model used the bearish version of the equity ranking system.
At this point in time, the super-bear conserve cash aspect of the model is active. That's because analysts are right now in the active process of cutting estimates on S&P 500 companies. This situation isn't forcing us to liquidate positions. But if we do close a position based on our normal sell rules, we would not reinvest the proceeds in new stock positions.
Here's more detail regarding each component of the model.
The Multi-market bull-bear ranking system
Actually, this is two ranking systems that are combined into one. Each group of factors, one bullish and one bearish, is weighted 50%. All factors are embedded within if-then statements:
- If bullish conditions apply, rank according to such=and-such criteria; if not, set the result to NA (which stands for “Not Available”).
- If bearish conditions apply, rank according to such-and-such criteria; if not, set the result to NA.
At all times, conditions will be either bullish or bearish. That means 50% of the system will produce numeric scores, and the other 50% will be NA.
The bullish system consists of three equally-weighted groups of factors: Growth, Value and Sentiment. Growth looks at recent (latest-quarter and trailing 12 months) EPS growth trends, and EPS acceleration (which compares the latest quarter growth rate to that of the trailing 12 months). Value looks at Earnings Yield, Price to Book, Price to Sales, and Enterprise Value to EBITDA. Sentiment looks at estimate revision, analyst ratings and changes in analyst ratings.
The bearish system has two equally-weighted sets of factors: Quality and Stability. Quality looks at returns on capital, financial strength, earnings quality, and longer-term EPS growth trends. Stability looks at Beta, Beta relative to Industry Beta, the range of analyst estimates, and the stability of estimates.
The market timing model
As noted, this is based on the Portfolio123 approach.
Generally speaking, the model is bullish if the risk premium on the S&P 500 (earnings yield minus 10-year treasury yield) is at least one percent, and if the 5-week moving average of the consensus current-year S&P 500 EPS estimate is above the 21-week moving average.
The specialized defensive-stock universe
Tight eligibility rules govern bear markets (defined as per the above timing model). Before a stock can be ranked, even using the bear-market rank, it must qualify for inclusion in the defensive universe. To be eligible, the stock must fit into one of the approved business categories, and sometimes, it needs to go further and satisfy some additional tests.
- Beta must also be below the utility average.
- Major Drugs
- Beta must also be below the utility average. Also, trailing 12 month sales growth must exceed the industry average and the Current Year EPS estimate must have gone up in the past four weeks.
- The company must also have generated positive Free Cash Flow in the trailing 12 months, the latest fiscal year, and the fiscal year before that, and the Price-to-free Cash Flow ratio must have been below the biotech industry average.
- Healthcare Facilities
- Beta must also be below the industry average.
- Medical Equipment
- Beta must also be below the industry average.
- Consumer Staples
- Waste Management
The super-bear timing rule
In the normal market-timing model, I evaluate S&P 500 EPS estimates by comparing a five-week moving average with a 21-week moving average. Notice that this alone does not require that estimates be stable or in an up-trend. To accomplish that, I add a super-bear protocol that prohibits new investment in equities if the latest current-year or next-year consensus S&P 500 EPS estimate is lower than its respective 10-week moving average. That would occur in a pretty stark environment and I wouldn't expect it to be in force often. As noted above, much of the bear-market protection comes from the specially dedicated rank and universe. For the record, though, the super-bear condition is active as this is being written, so proceeds form any current sales will be held in cash.
A more stringent bear-market buy-hold threshold
Under bull-market conditions, stocks may be bought or held if their ranks are 80 or above (100=best). As a practical matter, the buy threshold is not likely to have much impact. Since I'm only looking for 10 stocks, it's seems inevitable that I'll fill the portfolio long before dropping that far down in the ranking hierarchy. That may also be the case for bear markets, where the threshold rises to 90.
The difference is in the hold-versus sell decision, where stocks ranked 80-89 would have to be jettisoned under bearish conditions. The presence of this more stringent rank requirement makes it easier to clear space for defensive-universe/rank selections when market conditions change.
Different diversification rules for bull or bear markets
Diversification is classic risk-control strategy which, over a prolonged period, is more likely to reduce average returns but hopefully, result in a disproportionate reduction in volatility by cutting exposure to sector-specific factors.
During bullish periods, I decided to forego this requirement. At present, it seems to me that the regime-switching nature of the strategy is providing sufficient risk control in this ten-stock portfolio.
However, during bear market conditions, I decided to add the extra layer of risk control by limiting sector exposure to 50 percent of the portfolio. That threshold is higher than one might normally want for such a rule. But it's appropriate, perhaps necessary, here since the defensive universe that's active at such times already leaves us with so few sectors from which we can choose.
Thinking out loud
As noted, this model is offered for purposes of evaluation, to watch how it works in real time and see where it needs to be refined. Right now, I'm most comfortable with the basic market-timing model, the multi-market rank, and the defensive universe. That doesn't mean I'll never change them or add new variations. But it does mean that at this time, I consider them as much good-to-go as any other models I've used.
It's the latter aspects that will be evaluated most closely; the super-bear strategy (not so much the idea of having one, but the specifics of the rule), as well as the protocols regarding rank threshold and diversification. Other noteworthy areas for continuing study are portfolio size and rebalancing frequency.