SP500 Weekly Trend Following: Preliminary Analysis.
There are of course many different ways to invest and one method involves using weekly data. In this study, the price and volume data from the SP500 index was used to determine if there are benefits to using Friday's closing price in some manner to initiate a trade for the subsequent week so that 1% could be earned. In search of a winning strategy, several rules were created to define specific buy and sell signals which were ultimately designed to follow the week's trend. The week's trend is defined as the following:
In addition, if a Buy was indicated, then the trend was up and the number of times the market rallied 1% higher from Friday's close was tabulated. while for down weeks, 1% lower was used. Basically, if you could make 1% the following week then that week was counted as following the trend. Here are the results.
Before the results are presented, here is a description of the data sample. The most recent 15 years of weekly data was used in this sample: 9/01/89 - 9/17/04 which represents a total of 786 weeks of SP500 index data. As a reference, the SP500 futures started at 353.7 and rose to 1128.5, gaining 774.8 points, or 220%. So if you followed the rules of this study, the best any strategy could produce was 786%, but anything greater than 220% would be better then buying and holding. Second, the weekly volume figure was not a pure set. The volume figures used originated from the NYSE daily volume from Sept. 1989 to Nov. 2001 and SP500 daily volume was used thereafter. Third, this study used the average daily volume for the week rather than the total weekly volume. The reason for using the average daily volume for the week was because comparing volume between full 5 day weeks to holiday weeks consisting of only 4 days would be misleading and erroneous. And if one took more time, you could even produce holidays weeks with only 3.5 days. But since this is just a preliminary study, half days are considered full days in this sample.
Next, the results are presented in a table that allow you to easily compare the results from one strategy to the other. Each strategy's rules generated both Buy and Sell signals which were tallied individually and then as a group. The total wins and losses columns combine the data from both the Buy and Sell signals while the net column represents the difference between the total number of wins to losses. The "% winners" column equals the total wins divided by the combined total of wins and losses and the Chi Squared values represent the degree of non-randomness, and determine if these results occurred by chance or not. All values less than 6.64, indicate that the results are no different than chance, or flipping a coin. All values highlighted in green indicate that they are not random. In addition, when comparing Chi, those strategies with the larger Chi values increase your odds of success.
Rule 3 is of particular importance as it tested a popular and widely accepted strategy on Wall Street. It's always fascinating to discover how the most revered axioms on Wall Street don't pan out and it's equally remarkable to demonstrate that buying higher prices on increasing volume proves to yield the most unreliable results from week to week (rule 3). As a matter of fact, if you were to follow any one strategy, buying every Friday was simpler and better than all other strategies but one.
Buying higher prices with increasing volume from week to week didn't improve your odds of success and equivalent to flipping a coin. Concomitantly, selling higher prices when volume decreased from week to week didn't work well either. Third, selling lower prices with increasing volume didn't work either. But buying lower prices with decreasing volume from week to week produced the best Chi statistics and the best overall "% winners" of 67%, which is equivalent to 2 wins for every loss. The problem though is that the number of occurrences was low - 185 out of 786 weeks which works out to be only 1 trade out of every 4 weeks. Even the net wins of 63 indicates that you would be far behind the market if you adopted this strategy. So despite rule 4's great "Buy" side statistics, it's still a loser in the broader context of comparing the potential gain of 63% to the market's 220% over the same period. Also worth mentioning is how badly all of the rules performed.
| Results: Weekly Trend Following | Chi Squared | ||||||||||
| Rule | Buy - Win | Buy - Loss | Sell - Win | Sell - Loss | Total Wins | Total Losses | Net | % Winners | Buy | Sell | Total |
| 1 | 228 | 211 | 189 | 158 | 417 | 369 | 48 | 53% | 0.6 | 2.8 | 2.9 |
| 2 | 247 | 209 | 185 | 145 | 432 | 354 | 78 | 55% | 3.2 | 4.8 | 7.7 |
| 3 | 120 | 107 | 109 | 103 | 229 | 210 | 19 | 52.2% | 0.7 | 0.2 | 0.8 |
| 4 | 124 | 61 | 94 | 68 | 218 | 129 | 89 | 62.8% | 21.4 | 4.2 | 22.8 |
| 5 | 244 | 168 | 204 | 171 | 447 | 339 | 108 | 56.9% | 14 | 2.7 | 14.8 |
| 6a | 244 | 216 | 168 | 158 | 412 | 374 | 38 | 52.4% | 1.7 | 0.3 | 1.8 |
| 6b | 242 | 206 | 183 | 155 | 425 | 361 | 64 | 54.1% | 2.9 | 2.3 | 5.2 |
| 7 | 486 | 422 | 351 | 313 | 837 | 735 | 102 | 53.2% | 4.5 | 2.2 | 6.6 |
| 8a | 116 | 114 | 119 | 111 | 235 | 225 | 10 | 51.1% | 0.0 | 0.3 | 0.2 |
| 8b | 89 | 52 | 109 | 88 | 198 | 140 | 58 | 58.6% | 9.7 | 2.2 | 9.9 |
| 9 | 205 | 166 | 228 | 199 | 433 | 365 | 68 | 54.3% | 4.1 | 2 | 5.8 |
| 10 | 456 | 330 | 456 | 330 | 126 | 58% | 20.2 | 20.2 | |||
| 11 | 409 | 377 | 409 | 477 | 32 | 52% | 1.3 | 1.3 | |||
|
Note: Chi squared values
greater than 6.64 (1% 1 degrees of freedom) indicate non-random
events and are highlighted in green. Chi squared represents the
experimental control for this study which is identifying the odds of the
results being due to something other than chance. Other experimental
controls included are rules 10 and 11 which identify statistical streaks
or "runs". Notice that if you followed the rule of buying higher weekly closes with increasing volume (rule3), you would have generated 120 wins and 107 losses which is according the computed Chi Squared value (0.7) no better than flipping a coin. This low chi squared value indicates that the results are due to a random process and that this rule should not be used if you wanted to earn 1% during the subsequent week.
| |||||||||||
Now for a brief discussion of these results. Naturally, we want to find the best strategy but the question is how would you determine which is the best? The obvious choice is to find the strategy that produces the most money. But as we'll explain, none of these strategies did well as compared to the market's 15 year rise. Second, the table presented doesn't reveal which strategy made the most money, because there was no need to continue with this preliminary study as all of the strategies failed to pass the first step in the analysis.
For starters, seeing which of these strategies is better than buying and holding, is rather simple. A strategy must have a value greater than 220 in the Net column, as each win represents 1%. However please note that if a particular strategy were to have a net value of 220, it would not mean that the profitability of this strategy would equal the market's rise over the past 15 years. This is because the losses in this study were not limited to 1% and it was not determined if a loss of 1% occurred before a gain of 1% each week. These results merely show if it was possible to earn 1% and how frequently such as event was possible, and as you can see, no strategy reviewed in the table was better than the buy and hold strategy over the last 15 years.
However please note that this study is somewhat different from what Wyckoff traders do which is compare volume at previously traded highs and lows which are not limited to weekly highs and lows. The periods that they compare vary in time, while the time period in this study is rigidly set to one week. So in all fairness this is different. But if you were to use weekly data and weekly volume to bias your results, then you'd have to reconsider this approach as these data indicate that using weekly data doesn't produce any better results.
Next, let's review the number of winners. The "% winners" column has a range from 51%-62%. So all of them produced ratios that are similar with Rule 8a and 3 being the worst performers and Rule 4 and 10 being the best. Now this is extremely odd as Rule 10 blindly buys every Friday while Rule 3 only buys when the volume increases. One would think that since the 1990s produced the largest bull market in history, Rule 3 being a two dimensional strategy should be a better strategy than Rule 10 which simply always buys. Plus it is widely accepted that increasing prices with increasing volume is what is needed for the trend to continue. Well, in this study, it doesn't appear to demonstrate any immediate follow through or benefit to the subsequent week.
As in any study or experiment, controls are needed to demonstrate causality. Basically, you need a reference to make a judgement as to whether the idea being studied has any validity. So in this study, the following represent the controls. As in most financial studies, an idea must be compared to some reference such as a flip of a coin. One way would be to let a computer randomly generate buy and sell signals, but each run would yield slightly different results and many comparisons would need to computed. However, a faster and simpler method would be to subject the results to a "randomness" test. This is quite simply the Chi squared value. In this study, if the value of Chi squared is larger than 6.64 then it is more likely to be non-random process and has some validity. Any value less than 6.64 means that the results are no different than flipping a coin.
In addition to this randomness test, other controls were included to explore another characteristic of randomness such as "runs" or streaks. These are simply consisently buying or selling each week. This control doesn't test for randomness but explores how likely runs will occur and also form boundaries of performance, much like going to Las Vegas and playing dice and betting on 7 every roll. It shows the power of "runs" and their influence on the results. Clearly, the chi squared value presented show that the buyside has had a run, but interestingly, the sell side did not despite the recent decline from 2000.
So putting the % win and the Chi squared together, it's interesting to note that this study shows you that Rule 4 is better than Rule 3, which defies logic. According to these data, you are more likely to gain 1% selling Friday's close when it is below the previous week's close with increasing volume, than it is to buy Friday's close when it is higher than the previous week's close with increasing volume. It appears that higher prices with increasing weekly volume act as resistance while lower prices with increasing volume don't. ( while not shown here, this is opposite to how the SOX index behaves.)
Next, examine the Buys and Sells separately. Notice that there wasn't one case in which Chi squared proved the odds to be better than chance on the Sell side. Not one Sell strategy produced any remarkable strategies which gives a factual basis for why many investors should stay away from short selling. As for the buy side, the two strategies that produced odds better than 50:50, rule 4 and rule 8b, were the result of the negative case of the rule, buy if prices drop and volume drops. In other words; buy low and sell high; buy the dips. Another characteristic worth mentioning is that while the number of trades seems high, the results of all of those trades would be abyssmal. Your broker would be happy, but your dwindling account balances would make you unhappy.
So if you were hoping to make 1% per week on a consistent basis using any of these strategies, then your best bet is to simply buy each Friday. But if that's too easy for you, it is interesting to note that using the weekly midpoint is slightly better than using the weekly difference in the closing prices. Also, here's a "head scratcher". Rule 4 works more consistently than Rule 3 despite the fact that the 1990s was the biggest bull market in history! This is more frustrating given the fact that buying every Friday is definitely better than selling every Friday. So isolating volume as a variable appears to be make no impact on performance, and including it in your analysis doesn't appear to be advantageous on a week to week basis.
created 9/17/04, ©2004, The Small Investor's Software Co.