I’ve started to have some success with equity market timing and presented one such model on Seeking Alpha. So far, it’s been doing what I had hoped it would do. And now, the recent launch of ETF screening, ranking, and backtesting on Portfolio123 adds a whole new dimension to this endeavor, especially when I give some attention to the leveraged ETFs, which commentators love to trash. (“I thought you were supposed to double the market over a four week holding period – I closed my eyes when I got to your disclosures about the doubling being just a daily thing – but over the past month, you delivered only one and one half times the market’s return, so . . . Gotcha!”) As those who’ve seen my previous Seeking Alpha articles on leveraged ETFs know (click here, for example), I strongly disagree with the critics. Rather than playing silly “Gotcha!” games, I think it would be much more productive to learn how leveraged ETFs actually work, accept them as a new set of tools in the investor’s arsenal, and figure out how to use them, along with other tools, productively.
Market timing with stocks - great, but complex
The basic Portfolio123 market timing model is quite straightforward: be bullish if the S&P 500 Risk Premium is at least 1%, and if the 5-week simple moving average of the current-year S&P 500 consensus EPS estimate is greater than the 21-week moving average.
The overall track record of this model in its initial form, which assumes stock selection per a Portfolio123 Momentum rankings system under bullish conditions and all cash during bearish intervals (as defined by the timing model) speaks for itself, and quite well.
But the last month wasn't so great as the portfolio was all cash at a time when the market surged upward.
This isn't simply a matter of not-every-model-is-perfect-all-the-time. Instead, we're seeing something that plagues many successful models of all types, difficulty catching major turns. For this model to flash a bullish sign, we'll need generally rising S&P 500 estimates over the course of a 21-week period (enough for the 21-day moving average to beat the five-week moving average).
That's no small order. As of this writing, the 5- and 21-week moving averages of the S&P 500 estimate are such that if the estimate were to remain steady, it would take 20 weeks for the model to turn bullish. If the estimate were to jump to 30% next week (a very substantial bounce considering we're talking about an aggregate, not a specific company), we still wouldn't see a bullish signal till about mid-June, assuming, or course, that the S&P 500 risk premium stays above 1%.
It's tempting to simply revise the model. Perhaps 21 weeks is too much for the longer moving average. But look again at Figure 1. It's hard to argue with a track record like that. Shortening the moving-average period may work well now (assuming the market has passed its cyclical bottom). But it may not work over a prolonged time frame. As we try to adapt the model to force it to produce an answer we'd like to see get right now, we increase the overall danger of whipsaw, which is definitely not something we'd enjoy experiencing with real-money portfolios.
Another solution would be to give the timing model some help; add other factors that mitigate the impact of timing errors. I did that in the case of the model I previously described on Seeking Alpha, which keeps us in stocks even when the timing model is bearish. At such times, instead of going to cash, I switch to a conservative stock ranking system and apply it to a specially constructed universe of conservative stocks. (Click here for a Seeking Alpha article describing how this universe is created.) Figure 3 shows how it backtested over the past year.
It’s not beating the market. But coming this close to the benchmark on one of those rare occasions when the market rises more than 20% in a month; well, that's pretty good by most standards, especially if additional testing supports a hope that it would outperform over more prolonged time periods. The timing is still wrong, but the special universe and ranking system have softened the blow.
The negative, here, is complexity. Seriously! This is a very complicated thing to create.
Market timing with ETFs - a whole new ballgame
I used Portfolio123’s new ETF screener, including its if-then function, to create a simple screen which, in plain English, boils down to this:
- If the two market timing tests are fulfilled, select all ETFs contained in the “bullish” list I created.
- If either of the timing tests is violated, select all ETFs in my pre-stored “bearish” list.
From this point on, everything depends on which ETFs I put into these lists.
Figure 4 shows the results of a backtest of this screen when the bullish list contains only the ProShares Ultra S&P 500 ETF (SSO), and the bearish list contains just the ProShares UltraShort S&P 500 ETF (SDS). (That's all I need here. These are ETFs, not stocks. Just one will adequately diversify a portfolio.)
That's pretty good. But some might be uncomfortable with the jaggedness of the trend we see in the latter part of the period. So I fine tune the risk-reward profile to mitigate the impact of inaccurate timing calls.
Figure 5 shows performance if we take out the leverage within the ETFs. In the bullish list, I drop SSO and replace it with the S&P 500 SPDR (SPY). In the bearish list, I drop SDS and put in the ProShares Short regular short ETF (SH).
Figure 6 is a variation that would be preferable for someone who is basically bullish and wants to participate aggressively on the upside, while taking only moderate bets on the downside. The bullish list contains Ultra S&P 500 ETF (SSO), while the bearish list contains three ETFs: the non-leveraged short S&P 500 ETF (SH), the iShares 1-3 year treasury bond ETF (SHY) and the iShares 7-10 year treasury ETF (IEF). So now, the downside portfolio is two-third fixed income and one-third short equities (non-leveraged).
Over this particular test period, I was able to cut volatility without forfeiting good returns. Realistically, though, I won’t count on that happening over a longer time frame.
Figure 7 shows the results of a variation that's more consistent with the spirit of what I did in the above-mentioned equity model. In bearish periods, I'm not short or in fixed income. Instead, I'm long stocks but in a more conservative way. My bullish list has SSO, an ultra ETF, while my bearish list contains SPY and SHY (hence I'm using market timing to switch my equity exposure between 100% and 50%. Here's the backtest result.
This doesn't seem as great as some of what we saw above. But bear in mind this is just a three-year test, given the amount of time when I have data for all the necessary ETFs. For purposes of comparison, Figure 8 shows how the equity model performed over just the past three years.
Suppose I want to come even closer to the spirit of the multi-market equity model and look for conservative sectors during bearish periods. I could do that by using SSO on the upside, and for times when the model is bearish, I could go 100% into conservative equity sectors with the following: The Utilities SPDR (XLU), the Healthcare SPDR (XLV) and the consumer staples SPDR (XLP). Figure 9 shows backtest results based on equal bear-market positions in XLU, XLP, and XLV.
This is just a small sample of what's possible. Many additional variations can be created using standard (unleveraged long) ETFs, unleveraged short ETFs, double short ETFs and since late 2008, triple leveraged long and short ETFs from Direxion (with the latter, the downside is a short backtest period, due the very recent launch of these ETFs, but the passage of time will help a lot here). Moreover, leveraged and short ETFs are now available for many market segments, and sectors, and we're starting to see them come out for other asset classes, thus providing more opportunities for creativity.