My first question is whether volume affects range. In other words, I think that if we don't have many people trading on a given day, the market doesn't move. I have to test this: 1. Daily Volume related to Daily Range? (Constant relationship?) If the above hypothesis is true, question number 2 is if pre-open volume (on futures) and yesterday's volume can give us a good estimate of what the volume for the upcoming day will be. I have to test this as well. So: 2. Today's Pre-open volume related to Today's volume? 3. Yesterday's volume related to Today's volume? And last, "pre-open" range itself could be a good way to estimate today's coming range. Once again, we have to test it: 4. Pre-open range related to daily range? If we found these hypotheses to be true, then we could predict today's range, based on past volume (and/or range). A small range for the day would mean allowing only range trades (within support and resistance), and a large range would mean allowing only trend trades (beyond support and resistance).
q1. yes, volume and volatility are positively correlated. q2. yes, volume is autocorrelated. q3. same as (q2) above. q4. yes, volatility is autocorrelated. despite these relationships, their predictive value, which you are trying to figure out, is not easy to establish, at least not if you apply rigorous stats analysis. there are plenty of research papers on the subject. instead of re-analyzing the data, i have found my time better spent studying these. i attach an example of a paper documenting the volume / volatility / return relationship. all the best.
Thank you for your feedback and advice. I will look into the existing studies. However, it is so easy and fast for me to double check what these people found out, and see if it applies to the market I trade, that I am doing my own tests (I almost finished proving question number 1). Besides, that way I will devise specific rules for my own trading system.
Ok, so, I am reading the paper, which focuses exactly on my questions, and I am doing my own quick tests on the ecbot e-mini dow jones future. So far this is what I got (I won't get too deeply into all the details of my testing, sorry): daily volume positively correlated with daily range? YES, STRONGLY. volume of previous day positively correlated with daily range? NO EVIDENCE range of previous day positively correlated with daily range? YES, BUT TOO SLIGHTLY TO BE USED There still remains two tests, much more complicated for me, because they involve intraday data: 1. Is pre-open volume (ecbot e-mini dow jones) positively correlated to the day's volume (which in turn is correlated to range)? 2. Is pre-open range positively correlated to the day's range?
I finally got to the intraday range testing. This is something I have calculated from september 2005 to june 2006, and it might interest you. It's related to what we are talking about. It's the average % range measured dividing the day into half hours (Eastern Time): 23.59.00 0,03% 0.30.00 0,03% 1.00.00 0,03% 1.30.00 0,04% 2.00.00 0,04% 2.30.00 0,05% 3.00.00 0,06% 3.30.00 0,11% 4.00.00 0,09% 4.30.00 0,08% 5.00.00 0,07% 5.30.00 0,07% 6.00.00 0,06% 6.30.00 0,07% 7.00.00 0,07% 7.30.00 0,08% 8.00.00 0,08% 8.30.00 0,12% 9.00.00 0,14% 9.30.00 0,12% 10.00.00 0,30% 10.30.00 0,30% 11.00.00 0,26% 11.30.00 0,24% 12.00.00 0,22% 12.30.00 0,20% 13.00.00 0,18% 13.30.00 0,20% 14.00.00 0,21% 14.30.00 0,26% 15.00.00 0,24% 15.30.00 0,25% 16.00.00 0,26% 16.30.00 0,14% 17.00.00 0,07% 17.30.00 0,01% 20.30.00 0,04% 21.00.00 0,04% 21.30.00 0,03% 22.00.00 0,03% 22.30.00 0,03% 23.00.00 0,03% 23.30.00 0,03%
"9.00" means "the half hour that ends at 9.00 and begins at 8.30". In this sense, I made a mistake by placing "23.59.00" on top of the list, because it represents the last half hour of the day.
Ok, if anyone cares, here's my work. Range: Pre-open vs entire day: correl 0.48 Volumes: Pre-open vs entire day: correl 0.33
Looks about right ... no correlation. If you were to perform the same analysis during the Bubble you would get a completely different result. It was definitely much easier to make lots of money back then. I think it was around 2002 when all of the old reliable "easy money" relationships started to change.
Right, not correlated...I see. According to excel "correl" formula, it is slightly more correlated than these arrays which have a correlation coefficient of 0.39: 1 2 2 4 3 6 4 8 5 3 For me it's good enough to predict range 4 times out of 5.