I try very hard to capture specific phenoms for specific stock universes. One aspect to definitely include is making money all of the time. Dealing with all stocks is definitely something to avoid. One of the best aids to developing a model is to first include all the market variables in the model. Imagine what could happen to these curves presented here if one additional variable at a time were added.
<object width="560" height="340"><param name="movie" value="http://www.youtube.com/v/jxoqP6jQN34&hl=en&fs=1&"></param><param name="allowFullScreen" value="true"></param><param name="allowscriptaccess" value="always"></param><embed src="http://www.youtube.com/v/jxoqP6jQN34&hl=en&fs=1&" type="application/x-shockwave-flash" allowscriptaccess="always" allowfullscreen="true" width="560" height="340"></embed></object>
We're talking about a specific behavior here; trend following. The selection of your universe is just as, if not more important than the rules of the model. You've obviously never developed a model that works on all stocks. I suggest you keep trying rather than rationalize your way of doing things.
Then why doesn't your paper (attached) about the "p, v relation" make any mention of a specific stock "universe?" After I backtested it and you didn't like the results you tried to claim, after-the-fact, it should have been tested on a specific "universe." Why should something that broad not be a general concept that applies to all stocks?
thank you for responding. I agree that the selecetion of the Universe is the most important aspect of building systems going from models to development and then applications. From the beginning, 1957, I refrained from any generalizations to all stocks. I focused, instead, on making money and in particular the money velocity of making money. In other words I backed into the most important parametric measure of model building. The theories were not developed or writtien at that time. Now they are and they empower exveryone's thinking on how to go about getting the job done. Everyone rationalizes their decisions that give them the path that they follow. The merger of making money as fast as it is offerred became my basis for steeing up the foundation upon which I built. Both are easy to measure and then compare. The refinement of going from 30 cycles a year too 100 cycles a year in Trend Following was very interesting indeed. The Universe and the cycling within the Universe naturally leads to crossover trading based upon what is held compared to what is available, all based on money velocity. The unit of money velocity is well over 3.5% per day with roughly unlimited capital.
We all empathize with your difficulties. For anyone who wants to print the paper, please do so and reason how to use it to your best advantage. Think of the value a relation of Price and Volume has for stepping off into the greater picture. All tools work best when they are applied to the best opportunities. As Mike syas choosing the universe is very important. Trader666 take Mike's advice and go to the abundant resourses that are available for choosing the Universe that goes with the paper. Use the IBD data to select quality stocks to apply the paper to. Your back test showed no statisitical significance. That has been explained to you. If a person uses a technique on a large sample he finds out how segments of the large sample perform. You chose a sample of the 1000 stocks of the S&P 500. Most people would smile upon seeing the sample description as applied to the years 2002 to 2005. At any rate, it is a good idea to use the contents of the paper to define the way the backtest would have been done by someone who wanted to backtest a sample. You used spydertrader's coding to do the test, you say. Just to further empathize with you and your difficulty in getting any testing results (something statistically significant)I recommend using the other half of spyder's coding. You used his entry coding and you did not use his exit coding. If a person uses the coding for entry and for exit, then he finds out how long the "hold" is to complete a backtesting cycle. The answe is 8.1 days using 400,000 trials. You used the same hold for all entries. The hold was a "timeout" exit. In the paper the time to exit is symmetric with the entry: both are score changes and they are specifically cited: 0 to 7 for entry and 4 to 3 for exit. The scoring cycle repeats. This does not mean that there are equal days for each score. What is means, though, is that P,V ,A/D cycles and scores occur in a particular order and sometimes the score backs up occassionally before resuming its rotation. I have given you several aids to help you recover from your problems. A one pager on Unusual Volume, references to completed days of trading using the one pager based on the paper and years and years of posting of people using the method and posting their results over years. All stocks cycle. The range of cycle frequency is astounding. As compnies grow their cycling characterisitcs change. For people who trade, they use different criteria than those who invest. Trading 100 cycles per year is different than doing buy and hold. To use scoring on any stock it is best to select the cycle fractal of the stock and score it on that fractal. You chose to trade large cap stocks on a five day cycle. Your backtest proved that large caps stock to not have a statisitically significant price change over 5 days. These results are well known and well posted by you by now. For the sake of N00bies, were I you I would post these results periodically so that they can understand that trading large caps in position trading according to your backtest will not make or lose any significant money. Look at how many years you have waited to learn to make money position trading stocks. Certainly, you represent the majority of traders.
Oh please... I used your original document as the basis for my backtest. Read it carefully. On page 8 it says to buy the 0 to 7 turn of the "P,V relation." I used Spydertrader's code for the scoring so don't say I didn't do it right. And there are no conditions or filters saying which stocks to buy, so I tested it on 1000 S&P stocks over a five year period. I didn't choose "a sample of the 1000 stocks of the S&P 500." The S&P500 only has 500 large cap stocks in it. But the S&P1500 has 1500 stocks... small, mid and large cap. I've told you many times before why I used time exits instead of the 4 to 3 transitions called for in your paper. And the reason is, your "model" is so BROKEN that the transitions from 4 to 3 are VASTLY outnumbered by the 0 to 7 transitions... enough to make the average trade last for YEARS. So I exiting 5 days later (which worked better than 1,2,3, or 4 days). Time exits are a legitimate way to test entries. You can pretend they're not but that will only make you look even more ignorant about backtesting. Bottom line: Your "p,v relation" model as described in that paper is BROKEN.