Trends in financial instruments are extremely interesting. A certain trend-based model can provide consistent outsized returns in some applications, but when the same model is applied to a different market those outsized returns may vanish quickly. Likewise, similarly constructed trend-based models that vary only in duration of trends can have wildly different results on the same market.
On the one hand this may tempt us to create different trend models for different markets, but on the other we know it's dangerous to data mine results for a backtest. This type of period specific data mining can provide catastrophic results moving forward.
Over the past few weeks I have been looking carefully at complex trend models applied to a range of markets. This includes focusing on shorter-term trend models, longer-term trend models, blended trend models, long/flat applications to trend models, and long/short applications to trend models.
Trend Models on Equity Markets
When I want to do a long backtest on an equity market, I am generally confined to the S&P 500. This is the only market where I can get free daily or weekly data on that extends back to 1950. From there, I can get Nikkei 225 data from 1965, NASDAQ Composite data from 1971, FTSE 100 data from 1984, Russell 2000 data from 1988, and so on.
Unfortunately MSCI and FTSE keep a pretty tight lid on daily data from their more popular international indices: the MSCI EAFE, MSCI Emerging markets, FTSE Developed ex-U.S. and FTSE Emerging markets index.
Aside from being the most popular stock index in the world, this is why we see the S&P 500 being used as the example market in nearly every model published on the internet blogs. This includes my little writing project.
Long/Short Trend Model (Short-term)
Short-term trend following is very profitable in certain time periods. While it can appear highly technical and advanced, or at least difficult for your average self-directed trader to implement, it has become a lot easier with developments like the E-mini futures on major indices.
In this subsection I've simulated a long/short trend model on the S&P 500 (price only) using trends which are shorter in duration. I took more than 80 trend measurements ranging from 2.5 weeks through 3 months. These were derived into a factor ranging from -10 through +10. If the factor was -10 the investor was fully short; if the factor was +10 the investor was fully long.
A trader in the late 1980s who ran a long/short fund on S&P 500 going back to 1950 using a model like the one described above looked like a genius. It generally performed very well from early 1960s, but importantly it got you on the right side of the 1987 crash. Investors with any sense would have poured money into your hands.
Since 1987 things have not gone that well for this strategy applied to the S&P 500. The nature of the market seems to have changed after 1987 and the exact same model substantially underperformed the S&P 500 as seen below.
If you look at the poor performing chart more closely, there are moments of apparent brilliance. In September and October 2008, this single system would have returned nearly 40 percent. This is very aligned with the performance seen by some of the short-term trend equity funds at that time (often amplified by leverage).
Long/Short Trend Model (Long-term)
As one might predict, when we move to longer duration trends on the S&P 500 index the backtest becomes more stable. This is mainly because transitions from fully long to fully short occur more slowly.
In this subsection the trend durations used are much longer. Again I took over 80 different measurements of trend that ranged from 3 months through 12 months. I derived these into a factor ranging from -10 through +10. When the factor was -10 the investor was fully short the S&P 500; when the factor was +10 the investor was fully long.
Unlike the short-term trend strategy, this long-term strategy didn't have any long periods of outperformance relative to the index (except the period from 2000 until 2009 using peak-to-trough measurements).
The brief moments of massively outperforming the index were quite fleeting. This strategy saw a peak 55 percent jump in 1973-1974, a peak 45 percent jump in 2001-2002, and a peak 75 percent jump in 2008. On the surface this looks like amazing downside protection, but these short periods of outperformance were quickly followed by underperformance.
A quick spike up immediately followed by a quick drop is not a desirable or sustainable solution to portfolio construction. It's all but impossible to try ride the climb up only to get out before it drops down.
A rolling return average demonstrates the sharp moves in the long/short model relative to the index quite clearly. It also shows the sustained underperformance compared to the index.
Short-term vs. Long-term Trend Model
When we compare our short-term trend model to our long-term trend model, we can see a massive divergence and shift in overall market conditions post-1987. The long-term trend model began to outperform and just 10 years later had pulled ahead of the short-term model.
Applying complex trend models to an equity index like the S&P 500 can show some clear differences in how markets behave in short trends and longer trends.
We can see a pretty clear shift in how the S&P 500 behaved (and could be traded profitably) prior to 1988 and following this period. Before 1988, a short-term trend system that traded the S&P 500 both long and short was extremely profitable.
Since October 1987, short-term trends applied to the S&P 500 in the same manner would have performed very poorly. I don't have the necessary data to determine if the profitable period between 1960 and 1987 was an anomaly, or if the market shifted in nature after 1987. Either way, a short-term strategy that went long and short the S&P 500 after 1987 never again saw those great returns in any sustainable way.
Long-term long/short trend systems on equity markets can show impressive profits in downtrending markets. But it would be difficult to realize and hold onto those gains when the market reverses course. We saw this play out in 1973-1974, 2002, and 2008.
In future posts I will apply my trend model to long/flat strategies in the S&P 500 and other equity indices. I also hope to do the same with several of the major currency pairs.
Comments & Questions
All comments are moderated before being posted for public viewing. Please don't send in multiple comments if yours doesn't appear right away. It can take up to 24 hours before comments are posted.
Comments containing links or "trolling" will not be posted. Comments with profane language or those which reveal personal information will be edited by moderator.