| |
 November 2009 Sooner is Better than Later Our New Technical Indicators Not Only Beat the Benchmark, They Also Topped the Old Indicators
But all quantitative models are not fundamentally based. We also have a set of "technical" indicators that are updated weekly at the bottom of the Home Page. Unlike fundamental factors that are tied to a firm's sales, financing, and earnings, technical indicators focus on the trading patterns of stocks themselves. Like all financial assets, stocks' value is actually determined by supply and demand. Investors believe (hope?) that share prices are directly related to the underlying quality of the issuing company, and this may be true -- but only indirectly. What really determines if stocks go up or down on any given day is how many people want to buy versus how many want to sell. If the buyers outweigh the sellers, there's an increasing demand for shares and those who are willing to sell can ask higher prices. The opposite occurs when sellers flood the market. Presumably buyers appear because of fundamental reasons, but that's not always the case as was so clearly illustrated by the tech stock run-up leading off this century.
We have nine different technical indicators based on market volume, average daily highs and lows, the trends of two different indexes, and the advance/decline line. Each of these indicators gives a weekly reading of positive, negative, or neutral. When summed, these readings give the index a value ranging from +9 to -9. If it works well, following these readings should improve investment performance. This isn't to say that it's designed to beat a specific benchmark like the other quantitative models, just that it will improve results over a more haphazard approach -- the one followed by most individual (and some professional) investors. Our technical indicators have been around for more than fifteen years, going back to February 1994. Over that time there was the 1990s great bull market, the subsequent blow-up and the bear market of 2001 and 2002, the recovery, and now most recently, the bear market and (hopefully) recovery starting last spring. There have been a lot of buy and sell signals across that time, some have been better timed than others, but then this is the basis for a long-term investment approach. In the past fifteen years, there has only been once change made to the calculation of the index. In May 2002 the required reading for "sell" was lowered, enabling the index to make more timely sell calls. Was that a good choice? We were surprised by the result, and you probably will be, too.
Non-Symmetrical Simply put, there was nothing scientific about it. We knew the high readings (+9 and +8) should be buy signals while their counterparts at the lower end should be sells. But how far to go down the scale? And should buys and sells be symmetrical? In other words, should the top x numbers be buy signals and the bottom x numbers be sell? The goal was to add rigor to the investment process, so simple guessing wasn't sufficient. Many technical systems flash buy and sell signals daily, some even more often. We didn't want that. Our purpose was to design a set of indicators a long-term investor could use. While this wasn't buy and hold, we did hope to hold down needless trading and its associated costs. We didn't want new buy or sell signals every time the market sent a head fake.
Given all that, we decided to err or the side of too few rather than too many signals. Using the top two and bottom two numbers as buy and sell signals, respectively, seemed to make sense. Because the index was so limited, the market would have to be in a strong trend -- one way or the other -- to send a confirming signal (more on this in a moment). At the very least this should cut out short-term noise. With two numbers for buy and two numbers for sell, the results were symmetrical, too. There was no underlying necessity for this, but it just seemed right. Without additional data, there was little reason to make a buy decision quicker than a sell or vice-versa. However, additional data was available following the 2001-2002 tech meltdown, it was pretty obvious that the sell signals were too late to be effective. While it's often said that the best days of a market rally occur early in the period and missing just five to ten of them loses the majority of the gains, the same is true for market declines. This has been especially true in the past two decades when sell-offs have been sharp and relatively sudden. And investor holding cash will often fare better for a prolonged period of time relative to one who suffers a steep sell-off and holds on through the recovery. Being late to the rally doesn't hurt nearly as much if losses don't have to be recovered prior to building new gains. On the other hand, short yet sharp bear market rallies frequently occur in the midst of a longer term decline. These are sometimes called "sucker rallies" for good reason: Sudden market reversals can quickly wipe out short-term gains, leaving hapless investors in worse position than before. Based on the new data and the foregoing trading hypotheses, a valid argument could be made for an asymmetry between technical buy and sell indications. For the long-term investor, there was more benefit to selling sooner and buying later. This was incorporated into the technical indicators when they were reviewed in May 2002. Since then, sell signals occur at readings of -6 and -7, as well as the original -8 and -9. Buy signals remain unchanged, only coming at +8 and +9. In a sense, this was an odd change for a series of technical indicators which are usually viewed as market timing tools. In creating this set, we wanted to devise a tool that could be used by long-term investors, so risk plays a greater role. Adjusting the sell rating was a way to increase the downside protection, something not often considered in a short-term trading system. Now with over seven years of additional data, the change seems to be well supported.
Recent Results The nearby chart shows the results for the "Technical Portfolios" calculated using both the old and new interpretations, the S&P 500, 90-Day Treasury Bills, and the Barclay Capital Aggregate Bond Index. The Technical Portfolios are the results of trading between the three indexes based on readings from the technical indicators. Because these are based on indexes and not actual investments, results are gross because they reflect no management fees or trading costs. Nevertheless, an investor could closely approximate these results by using a discount brokerage and low-cost exchange traded funds (ETFs) for the stock and bond index along with a money market fund representing the T-Bills.
It's important to keep in mind that the technical indicators actually send three signals, buy, sell, and hold. The reading cannot go directly from buy to sell (or vice-versa) unless it first passes through at least one week with a hold rating. This was included from inception to prevent whipsaw effects that can occur in a turbulent market. Here's an example of how it works: Suppose the indicators have consistently been flashing buy signals. In this case, an investor following them would be fully invested in stocks. Under the new interpretation, when the first reading of -6 or less occurs, the indicators have moved to neutral and the investor should move into cash. A sell signal won't occur until a subsequent weekly reading of -6 or less occurs at which time the investor should move completely into bonds. In other words, after a buy reading, it takes two sell readings to generate a sell signal. Similarly, if, in the example, after moving to neutral from buy it would take two buy readings to create a buy signal. It's not unlike volleyball -- you can win the serve from the opponent, but you can only score when you're serving. The first change of reading simply wins the serve, it takes another to change the signal. Applying this to the data, as you'll notice from the chart, both the old and new interpretation of the technical indicators ran together until the end of 2004. They got off to a good start in 2002 as both were in sell territory putting them 100 percent in bonds as stocks continued to decline. They diverged in October 2004 when the new, more sensitive sell rating, took the new reading to neutral. As a result, this interpretation remained in cash for the nine weeks running from October 15, 2004 - December 17, 2004 while stocks continued to climb. Two subsequent buy signals took it back to stocks a that point, allowing it to track the old reading as a mirror image until late 2007. The difference between the two -- just under 7 percent -- representing the equity gain not captured by the new reading during the two months it was in cash. This is a good example of how one can fall behind by missing only a fraction of an equity run-up. Just looking at this period, one might conclude that the May 2002 changes were actually counterproductive. But the benefits became evident when the market melted down in late 2007. The S&P 500 hit an all-time high on October 9, 2007, but shortly thereafter, the subprime crisis took its toll on the markets. The new interpretation of the technical indicators sent a sell reading on November 2 and then a confirming one three weeks later on November 23. After only two weeks in cash, it was fully invested in bonds. The old interpretation didn't move to neutral (and the safety of cash) until February 29, 2008. The confirming sell signal finally came on October 3, 2008, almost a year after the new interpretation. In the meantime, bonds climbed 13 percent while stocks lost a crushing 42 percent. This more than wiped out the advantage the old interpretation established in 2004, leaving it significantly behind the new. Through the end of September 2009, both continued their sell readings. You'll notice from the chart that both closely tracked the bond benchmark for the past year. On October 2, 2009, (beyond the data set of this study) both moved to neutral and then on October 9, went to buy.
A (Large) Grain of Salt However, despite the late sell signal, the old technical portfolio was still up 78 percent, far better than the top performing index. This suggests that the "market timing" of moving from one asset class to another based on the technical indicators is better than simply buying and holding either one or a combination of the three benchmarks. Who says a good quantitative model can't beat the indexes? But this is far too strong a conclusion. As with all performance "mountain" charts, this one has to be interpreted in the correct context. For example, like all similar presentations, this one is very period sensitive. It helps immensely that both technical portfolios were already tracking the top performing index at the beginning of the period. As mentioned earlier, it's much easier to post gains over a longer period if one doesn't first have to recover from initial losses. Here, both technical portfolios got a big head start over the first fifteen months. Had they both started at neutral (much less buy), the results wouldn't have been so impressive. In addition, neither portfolio has really performed all that well while the market was recovering in 2009. It wasn't until after the measurement period ended that they issued buy ratings. Certainly there's something to be said for caution, but in this case, they've sacrificed almost a 50 percent gain over the past six months to achieve it. Put another way, the new technical portfolio lost approximately 30 percent of its lead over the S&P 500 from March 2009 - September 2009. What would you conclude if the test period had started in March 2009?
One thing this does suggest is that the May 2002 move to an asymmetrical index may not be what's really needed. Increasing the readings connected to sell signals helped avoid big losses, so maybe the same range needs to apply to the buy signals. One would think this would have been especially helpful in getting the technical portfolios back into stocks earlier when the 2009 rally began. Looking back, it would certainly have helped. If buy signals included +6 and +7, the initial move into neutral would have occurred on July 31 and the confirming buy signal would have come two weeks later on August 14. This would have gotten the technical portfolios into stocks about a month and a half sooner. This clearly wouldn't have captured anywhere near the return of the S&P 500's strong March - September rally, but it would have gotten more of it. Over the entire measurement period, there would not have been any negative implications of using this interpretation other than the fact that it would have issued 54 more buy readings over the 397 months. Most of these would have occurred during the periods when the new technical portfolio was already rated buy, so they would simply be ongoing confirmation. In essence, the late summer 2009 buy rating was the only material impact. Because of this, effective October 1, 2009, we are incorporating this change in the technical indicators. Once again the reading is symmetrical: -6 to -9 are sell readings, while +6 to +9 are buy readings. We believe this is justified because of the negative effects of too slowly returning to equities when the market heads up plus the fact that it would not have led to premature changes during the volatile bear market in 2008. As time goes by, we'll check back to see how it's working.
Given this analysis' period-dependence, you might wonder why we didn't go all the way back to the technical indicators' 1994 inception to test the results. While that certainly would have provided greater and more diverse time period, it also would have violated one of the major tenets of good statistical analysis known as data mining. One of the biggest mistakes a quant can make is to create a theory based on a data set and then test that theory on the same data. Inevitably, the results will be wonderful, and who would expect otherwise? If you study a period to see what works, make a model based on it, and then test the model in the original data set, of course it works! It has to, so you haven't substantiated anything. Because the May 2002 changes were based on observations from 1994 to May 2002, the only real "out of sample" data available is from that time forward, the period analyzed above. That's also why we'll test the new changes from October 1, 2009 forward. Search this site! Just enter you key word or words: Get current quotes or follow your own custom portfolio,
courtesy of E-Line Financials:
|
|||||||||||||||||||||||||||||||||
UR QUANTITATIVE EQUITY MODELS are based on fundamental factors such as Price to Cash Flow, the PEG Ratio, and Debt to Capital. To create them, we used ten-year statistics from the period ending December 31, 1999 to create stepwise multiple linear regressions. That sounds impressive, but what it really means is we started out with ten year returns and eight other fundamental variables. The process eliminated any variables that weren't statistically significant (i.e. had no meaningful impact) to produce an algorithm combining the surviving factors' weights and values to produce a forward looking return projection. (If you want more details, you can find them 
