Do you like what’s hot or do you like what’s not? Put another way, are you a momentum player or are you a contrarian? Practitioners of technical analysis would phrase it like this: Do you prefer trending systems or those based on oscillators? Equity investors have long pondered such questions. But nowadays, with the ETF universe having become as large as it has, it becomes important here too.
Perhaps Mr. Market can help us decide which type of strategy we should pursue for a particular stock, or in the case of this article, ETF.
Listening to Mr. Market rather than laughing at him
A long-cherished segment of investment lore assumes that the Street as a whole is ignorant. The most famous example is Ben Graham's parable about Mr. Market. He is the archetypical supposedly manic-depressive investor who one day offers to buy your shares at ridiculously high prices only to offer laughably low bids the next day. Contrarian strategies based on trading against Mr. Market are followed by many, and such rhetoric is often pushed by pundits who relish the rebel role.
In the new ETF section of Portfolio123.com, I created what I refer to as an Adaptive Oscillator. I think of this model as the anti-Graham. Instead of making fun of Mr. Market, I sort-of hire him to serve as Director of Investment Strategy. He is going to say, ETF by ETF, whether I should be using a trending strategy or looking for a reversal.
The way this is done is actually quite simple and is based on the very nature of oscillating indicators.
Different oscillators use different formulas but generally speaking all attempt to measure a stock's price momentum and translate it to a score ranging from 0 (worst) to 100 (best). Practitioners usually assume scores above 70 indicate shares are overbought while scores below 30 signify oversold conditions. Not surprisingly, overbought stocks are to be sold or avoided, and oversold shares are considered Buys. I decided to work with the Ultimate Oscillator by Larry Williams but changed the thresholds to 60 and 40 respectively, to put larger numbers of ETFs into the extreme groupings.
Consistent with tradition, I assumed that oscillator scores above 60, suggested overbought conditions. But instead of assuming Mr. Market is acting crazy, I decided to follow his advice. I'm assuming that nowadays, when anybody with access to a free web site can see all the oscillator scores they want with just a few mouse clicks, if the Street is going to tolerate an "overbought condition," there's probably a good reason for that. So in my model, stocks with scores above 60 will be ranked based on oscillator scores with higher levels being preferred. I add a second trend-oriented requirement, a sort from high to low based on 5-day exponential moving average minus 20-day average, just to enhance the likelihood that investors are still putting their money where Mr. Market’s mouth is (i.e. I don't want overbought ETFs that are losing momentum).
Conversely, if the oscillator is below 40, I recognize the oversold condition and the potential for a bounce. So for these ETFs, I reverse the ranking such as to make lower scores preferable. That raises the question of where Mr. Market, my new strategy director, fits. One would think he was most likely to have been guilty of lowball selling. Actually, Mr. Market is contributing to the second component of the oversold-ETF portion of the model. I also rank from high to low based on 20-day exponential moving average minus 20-day simple moving average, thus looking for some market-based indication that the ETF is starting to correct out of its oversold condition.
As to middle-ground situation (the majority), where oscillator scores are between 40 and 60, I ignore the oscillator altogether and simply rank the ETFs on the basis of 5-day exponential moving average minus 20-day average. This does add a trending bias to the overall model, but I can live with it given that it seems consistent with my overall experience and that backtested well (see below).
This three-part if-then ranking system is run against a list defined by a screen that eliminates HOLDRS (which only trade in 100-share lots) and municipal bond ETFs, those whose average volume over the past 20 trading days is below 5,000 shares, and those which are based on short or leveraged strategies.
Figure 1 shows the result of the initial backtest. It assumes a four-week rebalancing period, prices based on the average of the next day's high and low, and consideration of the top ten stocks.
On balance, the model outperformed the S&P 500 for most of the period up till the financial crisis. Like many models, the recent past was challenging, although interestingly, this one's worst period of relative weakness occurred early in 2008 ratter than in the latter part of the year, by which time, it was doing a reasonable job pointing investors away from basic U.S. equities. The last few months have been especially strong.
Note that most ETFs will fall in the middle ground, with oscillator scores ranging from 40 to 60.
How much of the performance that we see above is based solely on the efficacy of the selection protocol we apply for that group, and how much benefit really comes from applying an adaptive oscillator to the upper and lower end of the range? Figure 2 will help us get a sense of this. It repeats the above test parameters but assumes the middle-ground selection rule (rank based on 5-day moving average minus 20-day average) applies across the board.
Beauty is in the eye of the beholder, and so too are comparisons like this. However, I feel comfortable assuming most observers would find the Figure 1 result to be better. Table 1 helps us quantify this by comparing four-week average excess returns from each model.
Average 4-week Performance Portfolio minus S&P 500
We see that the strategy's adaptive qualities had a negligible impact in down markets (strictly speaking, it's nominally worse), but that it contributed noticeably during up periods.
By the way, the result we see in Figure 2 supports my decision above to introduce a trending element even to oversold situations.
Using such a model
As a pure idea generator, we’re good to go. Look at the top ten ETFs and make choices from the list, which is as follows (in descending rank order):
Market Vectors Indonesia Index (NYSEARCA:IDX)
SPDR Dow Jones Small Cap Growth (Pending:DSG)
Fidelity NASDAQ Comp. Index Trust (NASDAQ:ONEQ)
United States Oil Fund LP (NYSEARCA:USO)
United States 12 Month Oil Fund (NYSEARCA:USL)
SPDR S&P 500 (NYSEARCA:SPY)
United States Gasoline Fund (NYSEARCA:UGA)
iShares Morningstar Small Growth (NYSEARCA:JKK)
SPDR S&P Emerging Latin America (NYSEARCA:GML)
iShares Russell 2000 Growth (NYSEARCA:IWO)
We’re definitely getting thumbs up with small-cap growth and energy as well as bits of developing-market general-market exposure.
In the ETF world, many funds with similar objectives are likely to be highly correlated. Don't expect such correlations to equal 1.00. There will be some portfolio differences, as well as differences in expense ratio, liquidity, etc. But it may well be that those differences are not worth pursuing if we can save trading commissions on a regularly rebalanced portfolio.
Assuming I'd like to have one fund from each likely-correlated subgroup, and that I'll choose the one with the highest average 20-day trading volume, here's what a reduced list might look like.
This certainly isn’t the last word on the topic. I plan to continue to refine this approach (obviously, the more depressed portion of the list needs more attention. But even now, I note two interesting points. First, there is something to be said for assuming that if an ETF is significantly overbought, there may be a good reason for that. Second, while ETF investors tend to be very style oriented, there’s something to be said for approaches that simply scour the world, this actually being the second model I created along these lines (click here for a Seeking Alpha article describing the other approach).
Disclosures: No positions