August 3, 2014

What follows is an experiment in strategy design.

Scenario: from one stock, over an 8-month period of the past year, I predetermined trade entry and exit points by date. Therefore, this experiment is entirely fabricated. Nonetheless, there was something to learn from the process.

Using one stock (AXP), I hardcoded trade dates and produced the following for summary performance report:

AXP - Hard Coded Dates (past year)

AXP - 1 year

(click to enlarge)

The above does show that all trades were profitable (100% hit rate). How could it be otherwise? The dates were selected from hindsight... So there was no surprise there.

Now for the experiment

From the above, what was kept as a trading strategy was the entry and exit dates for all the AXP trades. If these dates were applied to other stocks, they might or might not produce a winning outcome. Some might even go as far as saying a 50/50 chance of generating a profit or a loss on any of those trades.

Well, the outcome is somewhat different. I tried it over the last year of data on the 30 Dow stocks plus 2 that I've used in other tests. The output was:

DOW – 32 Stocks – AXP Fixed Dates

DOW Stocks - Fixed Dates

(click to enlarge)

The above summary performance report showed the expected 1,600 trades (32*50). They came in with an 89.69% win rate. Only 168 trades (10.31%) generated a loss, and even there, the average loss was only -2.41% per losing trade. Just one stock out of the 32 ended with a loss.

DOW – 32 Stocks – Fixed Dates

32 Stocks Fixed Dates

(click to enlarge)

This would imply that the dates chosen for AXP must have been quite general in nature in order to apply to about 90% of the 1,600 trades. And if it was that general, how about pushing the envelope further? Why not try this on another and bigger dataset: the Russell 1000 with its 985 stocks? What follows is the performance summary report for that one:

Russell – 985 Stocks – Fixed Dates

985 Russell 1000 Stocks

(click to enlarge)

This time, you had 49,100 trades, of which 40,965 (83.43%) were profitable.

Why do such tests?

The first strategy on AXP is idealistic, not the optimum, but surely over-optimized and certainly over-fitted. However, after having applied it to the 30 Dow stocks and the Russell 1000 stocks, a pattern emerges.

If one could design a trading script that could send its trading signals around the same date zones as in the AXP example, then one might have a winning trading strategy that could also survive in a much larger group as in the remaining 984 Russell stocks. One would not be interested in the trading signals per say but the trigger dates on this one stock (AXP) that could be passed on to other stocks in the portfolio as their trading signals. A one size fits all kind of thing.

Based solely on the above experiment, I would venture that AXP's price movements were somewhat representative of a much larger stock universe (984 + 30). And some of the reasons for AXP's profitability would be about the same as for the group as a whole.

AXP served as a bellwether stock. All trades in all the other stocks were copycat trades. A conclusion to such an experiment would be to note that a majority of stocks (90% in Dow stocks and 83% in Russell stocks) had about the same type of price patterns, the same general "trend" as AXP's; resulting in an exo-trading strategy with a positive and profitable "edge".

This would indicate that there was a general uptrend for AXP which could also be found in most of the 984 Russell stocks as in most of the 30 Dow stocks. Also, the path taken by AXP over the past year must have been highly correlated to the price movements of most stocks reviewed in this test. It all could be summarized in a single decision: trade AXP's general trend, copy trade date triggers to other stocks, and have fun.

What is left is to design a signal-triggering function that will mimic AXP's price movements and determine its entry and exit trading zones approaching the idealized model. I do have such procedures, so I look forward to testing them at some future time.

It was sufficient to use the trading dates on a single stock to win over a thousand. Interesting.

Created...  August 3, 2014,    © Guy R. Fleury. All rights reserved.