John P. Hussman, Ph.D.
One of the questions we often receive is why we don’t simply lift our hedges when the market advances above some moving average or another, and replace them when the market breaks below those moving averages. Certainly, when one looks a chart, extended market advances always break above various moving averages, and extended market declines always break below various moving averages, so simple trend-following strategies seem utterly self-evident. Unfortunately, if you actually take that strategy to historical data, the results typically aren’t nearly as compelling. Moreover, once any amount of slippage or transaction costs are taken into account, the most widely-followed strategies generally underperform a passive buy-and-hold strategy over time, and often don’t even manage downside risk particularly well.
I should emphasize up-front that the focus here is the use of popular moving-average crossover methods, and is not a criticism of trend-following methods more broadly. The second half of this comment discusses considerations that I believe are useful in evaluating market trends and extracting signals from financial and economic data.
The charts below feature a variety of moving-average crossover strategies using the S&P 500. The darker blue line tracks the total return of the S&P 500, and the others track a variety of strategies that are long when the S&P 500 index is above the given moving average, and in T-bills otherwise (the moving averages are exponential – giving a weight of 2/(N+1) to the most recent observation, and the rest to the prior value of the MA, where N is the length of the MA). The moving average lengths are Fibonacci numbers, which is a common practice among technicians.
At first glance, these strategies seem very promising, particularly for the 21-week moving average crossover, which produces a cumulative return about 40 percentage points more than a buy-and-hold since 1950. Unfortunately, over 62 years, that difference is actually quite small, amounting to just 0.5% annually. And even that excludes all transactions costs, taxes, and slippage. If you look carefully, you’ll also notice that drawdowns are not particularly small. In fact, the 21-week crossover strategy has a maximum drawdown during this period of nearly 40%. The reason is that the strategy has had an awful tendency to buy into rallies just at the point where they fail abruptly, and to sell into declines just at the point where they rally sharply. You can see this reflected in the drawdown of the 21-week crossover strategy (purple) during the 2000-2002 decline. The trend is your fickle friend.
A somewhat more realistic picture emerges if we consider even modest transaction costs. The chart below reflects a cost of 0.25%, which in today’s terms is the equivalent of allowing for the S&P 500 to slip about 3.5 points between the time of the signal and execution. Given tax differences between long-term and short-term gains, as well as brokerage costs, this isn’t an unreasonable cost to assume, but it is enough to entirely destroy the advantage of these crossover systems, particularly shorter-term ones that trade a great deal.
There are numerous ways to reduce whipsaws and trading frequency, for example, adding some “padding” around the moving averages and requiring the index to trade through the average by 1% or so before the signal is taken, but these filters also have a trade-off in terms of timeliness. Another way to reduce whipsaws is to track when one “fast” moving average crosses another “slow” moving average, which is the basis of MACD-type systems. Among the foregoing moving averages, the 21-week / 34-week crossover performs best, and very slightly survives even a 0.25% slippage, though it still experiences drawdowns of more than 25% in post-1950 data, and much worse drawdown losses in Depression-era data.
To provide a better view of the full period, the chart below presents the same data on log scale. Note that the combination of whipsaw losses (buying into advances just before they fail, selling into declines just before they rally) and transaction costs are most severe for the 8-week crossover, which has a maximum drawdown of nearly 45% in the 1950-2012 data. This is exactly opposite of what many investors assume. There is a belief that if investors use a sufficiently short moving average, they will “catch” emerging advances and will quickly identify market declines, allowing them to have their cake and eat it too. This is simply not supported in the data.
Does all of this suggest that trend-following measures should be ignored? Not at all – but it does emphasize that simple trend-following schemes (like moving-average crossover rules) are unlikely to be very effective in isolation, and are not usefully applied as a top-level filter in the hope of catching market rallies without being vulnerable to downside risk. Typically, the best that can be achieved with popular moving-average crossover systems is a moderate reduction in drawdown risk, but zero or negative incremental long-term return versus a buy-and-hold.
It’s also notable that since the 2009 low, and even ignoring transaction costs, the foregoing moving average crossover strategies would have lagged the S&P 500 by anywhere between 64 percentage points (13-week) and 104 percentage points (34-week). Including transaction costs increases this gap for all crossover strategies, with the 34-week crossover strategy achieving a cumulative total return of just 12% since the 2009 low (3.3% annualized). Clearly, moving-average strategies can have substantial tracking risk versus a buy-and-hold approach.
I am comfortable with the tracking risk that our own hedging strategy experiences because I believe that we can reasonably target full-cycle returns substantially above a passive buy-and-hold approach, with significantly reduced drawdowns. We’ve achieved both objectives in prior market cycles, measured from bull-market peak to bull-market peak, or bear-market trough to bear-market trough. But this was not the case in the 2007-2012 cycle, due to missed returns in 2009-early 2010 as we worked to make our approach robust to Depression-era outcomes. From a fiduciary perspective, I continue to believe that ensuring the ability to withstand extreme strains was necessary. From a practical perspective, I continue to believe that the ability to withstand extreme strains will be more relevant in the coming years than investors would presently like to believe.
So what is the difference between mediocre trend-following measures and more historically useful approaches? The key issue here goes back to what I’ve always emphasized about “signal extraction.” Generally speaking, single indicators provide weak information because the true signal (whether about market conditions or economic prospects) is invariably confounded by random noise. The ability to infer signals from noisy observable data is typically enhanced by using a variety of indicators. Nearly all modern signal processing – in fields as diverse as radar tracking, MRI scanning, and genetic analysis – is based on the observation that multiple sensors are best suited to picking up true signals in the presence of random noise. Just as we find for economic data, market action should always be analyzed in the context of multiple indicators that capture a broad range of sectors, security types, yield-spreads, leadership, and so on. The information isn’t just in the obvious trends, it is also in the less obvious divergences.
Ever since the late-1800’s, careful market analysts have been very aware of this fact as well. Consider Robert Rhea’s exposition of William Hamilton’s work in his book, The Dow Theory:
“The most useful part of Dow Theory, and the part that must never be forgotten for even a day, is the fact that no price movement is worthy of consideration unless the movement is confirmed by both averages. Many who claim an understanding of the Theory consider only the movements of the Industrial stock average if they happen to be trading in industrials. Some even chart only the one average and profess to be able to interpret the movements correctly. It is true that there are times when such conclusions seem justified, but over any extended period such procedure inevitably results in disaster… When the Averages disagree, it’s usually a sign of distribution.”
I’ll add as a side note that veteran Dow Theorist Richard Russell expresses enormous concerns today about market action, partly (though certainly not entirely) because of the joint breakdown in the Industrials and the Transports a few months ago, and the subsequent failure of the Dow Transports to confirm the rally in the Dow Industrials since then. Price-volume behavior is also problematic here. As William Hamilton observed a century ago, a market that becomes “dull on rallies and active on declines” should not be trusted. From that standpoint, the wholesale collapse of trading volume in recent weeks is almost creepy.
Based on a century of historical data, the multiple-sensor approach is what turns out to be most effective in our own work. In that sort of analysis, concepts like “breakout” and “crossover” turn out to be far less useful than concepts like “uniformity” and “divergence.” While our broad measures of market action have been unfavorable for some time as stocks have played hot potato between periods of overbought froth and periods of ragged market internals, we now see an unusual combination of both. Even the best trend-following measures we track broke down in mid-April as a result of clear deterioration in market internals (see No, Stop, Don’t), and we have not yet observed a recovery on that front.
So for those who have asked whether we can reduce the extent of our hedging until the trend-following components of market action turn negative, the simple answer is that they’ve already done so. Moreover, I should note that even very popular trend-following approaches such as the 200-day moving average, the “Golden cross” (50-day vs. 200-day), the 34-week crossover, the 55-week crossover and others, have produced flat or negative total returns – even before transaction costs – since the April 2010 market peak. Saying that some trend-following measures are “positive” is much different than saying that they are promising.
Still, it’s worth repeating that we addressed a more general trend-following issue earlier this year, in order to reduce our use of actual put options in a world where monetary policies make investors believe that free ones can be taken for granted. Consistent with the analysis above, historical tests indicated that it would be unprofitable to simply remove hedges whenever the trend components are favorable. A strategy like that has no exit criteria other than a trend breakdown, and when stocks are severely overbought, the required decline can be very steep and incur a great deal of loss before establishing a hedge.
Still, that fact suggested its own solution, which was to allow sufficiently overbought conditions to act as an alternate exit criterion. That approach turned out to be very effective, particularly in reducing the frequency of “staggered strike” hedges without reducing their long-term benefit. Accordingly, we added criteria earlier this year to restrict the use of “staggered strike” positions, requiring not only a very negative return/risk estimate, but also either negative trend-following measures or the presence of hostile indicator syndromes (e.g. “overvalued, overbought, overbullish” conditions). The limited set of instances that survive those criteria are historically associated with average market losses on the order of between -25% to -50% at an annual rate, depending on the particular set of syndromes involved. There’s not a chance that I would ignore that data here.
While our estimates of prospective market return/risk are presently in the most negative 0.5% of historical data, based on horizons from 2-weeks to 18-months, it is easy to blend our present defensiveness into our stress-testing period in 2009-early 2010 and simply assume that market conditions here are really no different than those that have generally accompanied the market’s run since the 2009 low. This would be a mistake. The phrase “most negative 0.5% of historical observations” reflects both a large magnitude and a low frequency.
In my view, the period since March 2009 can be reasonably divided into three segments; one generally favorable, one moderately unfavorable, and one extremely hostile. The first of these runs from March 2009 to April 2010, that end-date being the last time that our investment strategy would have accepted significant market exposure, based on ensemble models we developed during our stress-testing period, and which are presently in use. This segment began in March 2009 with our estimates of prospective 10-year returns above 10% annually, but with other features sufficiently characteristic of Depression-era outcomes that stress-testing against that data was necessary.
The second segment runs from April 2010 to early March 2012, when our present methods have been – with a few small exceptions – persistently defensive. During this period, we did spend more than we would like in hedging costs due to put-option decay, and have added additional criteria (related to trend-following measures) that would have reduced those costs without eliminating the intermittent benefits we’ve had from those positions. But this was a minor change compared to the introduction of the ensemble methods, and addresses what was essentially just a periodic inconvenience due to central bank interventions. Our investment stance would have been largely defensive since April 2010 in any event.
The third segment is the period from early March 2012 to the present, where our estimates of prospective market return/risk have been extremely negative on intermediate horizons out to 18 months, where our estimates of 10-year prospective returns have hovered around 4.5% annually, and where there are sufficient divergences in market internals, overvalued-overbought-overbullish conditions, and evidence of exhaustion to motivate a strongly defensive stance. In March, these estimates dropped into the worst 0.5% of historical observations. In mid-April, even our best trend-following measures broke down, and have not recovered because of broad divergences in market internals during the recent bounce. This set of conditions survives the tight restrictions required of our most defensive investment stance. This isn’t a stance that we expect to maintain indefinitely, but I can’t emphasize strongly enough that investors would be wrong to simply blend present conditions into the longer period since early-2009 as if there is no distinction.
That said, the next few weeks will include significant “headline risk” for the market, as Fed Chairman Bernanke speaks at Jackson Hole on Friday, the ECB meets on September 6, and the German Constitutional Court rules on the legality of the European Stability Mechanism on September 12. Accordingly, it is important to inventory our risks:
In Strategic Growth Fund, we have two primary risks. The first is the risk that our generally defensive and value-conscious portfolio of individual stocks will lag the indices we use to hedge, either falling more in a market decline, or advancing less in a market rally. This is always a risk that we accept when we are hedged. While our stock selections have strongly outperformed the indices we use to hedge since the inception of the Fund, there are certainly periods where that doesn’t occur. We’ve focused a great deal of attention on maintaining portfolio characteristics that we expect to perform well over time, but it is reasonable to point out that we remain light on financials and cyclicals, as global credit risks and recession concerns loom. The other risk is the “staggered strike” position of the Fund, which represents just under 2% of assets in put option time-premium looking out to the end of the year. I expect this position to provide significant hedging benefits in the event of a significant market decline (particularly in the event of indiscriminate selling), but we’ll also experience some time decay in this position if the market remains stable or advances over the next four months, which is not our current expectation.
In Strategic Total Return, our main risks are reflected in the 10% exposure that the Fund carries in precious metals shares, and in the Fund’s duration of about 1.8 years in Treasury securities. Though the ratio of physical gold to gold stock prices (based on the XAU) remains well over twice the historical norm, which provides some margin of safety, a strong commodity selloff could put pressure on gold stocks, and a 20% decline in those stocks, for example, would translate to a Fund decline of roughly 2% driven by our precious metals shares. Meanwhile, a 1.8 year duration essentially translates into a portfolio change of about 1.8%, on the basis of bond price changes, for every 100 basis point move in interest rates. Barring a very strong advance in interest rates, I view this source of risk as fairly muted, but again, I believe that present economic and market risks make it important for investors to inventory every exposure.
In Strategic International, our primary risk is the potential for differences in performance between the stocks held by the Fund and the indices we use to hedge. Here again, we’ve focused our attention on maintaining a diversified portfolio with characteristics that we expect to perform well over time, and both our holdings and our hedges are geographically diversified and reasonably matched, but we remain light in financials and cyclicals, which is likely to be a source of some day-to-day divergences.
Finally, in Strategic Dividend Value, our primary risk is that the Fund’s most defensive stance is a 50% hedged position, so while dividend-paying stocks tend to have lower sensitivity to market fluctuations, we estimate that the Fund has a moderate net sensitivity to market fluctuations even given its partial hedge. I believe that Strategic Dividend Value is suitable for investors who are bringing their market risk down from a more fully invested stance, but the market environment remains hostile enough that my preference remains for shareholders to defer moving from risk-free investments to the Fund just yet, since I do expect that the Fund will have some amount of exposure to market losses.
In short, we understand our risks, and we believe that they are acceptable in view of the prospective returns that we associate with them at present. My concern is that in the face of very widespread complacency about global economic recession, and the nearly infinite faith in the redemptive capacity of monetary policy, investors have neither taken an inventory of their risks, nor have any sense of the low prospective long-term returns (and potentially awful intermediate-term returns) that are presently associated with those risks.