Did quickie autocorrelation study of the H-L range. It suggests that: Real: 3, 5, 8, 9, 10 Random: 1, 2, 4, 6, 7 I'm least confident about #4 and #10. BTW, If you want to create more "realistic" random price data, I'd suggest some sort of ARCH or GARCH model to add in autocorrelated volatility. Thank you for a most interesting challenge, I look forward to the answers and further challenges. Trade well, Traden4Alpha
impressions based on visual inspection of candle charts only, no fancy analysis: 1 random 2 real 3 random 4 real 5 real 6 random 7 random 8 real 9 real 10 random
my GOD man! if you are plannng on doing something like this, try a little harder, or don't post the CSV file. it's a dead simple thing to figure out when all the starting prices are the same if i'm wrong, then please ignore this post /j http://toronto.tasug.com
LOL!!! I did this on purpose! I transposed all the real data to start at 30! If I left the real prices it would be too easy to figure out which is which. Trust me, they are NOT all random. You will see when I give the results. I will reveal which charts are real and their exact dates, so you can check for yourself.
After this round is completed, we might want to try again using around 1000 bars of data instead of just 300. We have the possibility that some real charts look very random, and some random charts look very real, when looking at a smaller sample size. I'm also curious what people's reactions are to the "real vs. random" when there is more data to look at. Here are some URL's on random walks for those interested:   http://reallifemath.com/randomwalk.html   http://scienceweek.com/2001/sw010622.htm   (scroll down to section 2 about middle of page)
I seriously thought about this, but I find a rather significant problem, the real data. If it is a long history, it is MUCH more likely that the average trader will recognize it, even if only subconsciosly. Maybe an idea would be to use actively traded foreign stocks? I do like your suggestions for a more scientific contest in the future; Using brownian motion or something of the sort. But like I said, the most serious obstacle is choosing the real data... If you choose a long history the experienced trader would recognize it. But if you only count on the un-experienced trader the results are also meanignless because they obviously have no skills reading the market. Any other ideas around the problem?
You could use intraday data, using like 120m bars, or something a little more obscure like 111m bars. Also since you've normalized the data so someone couldn't simply do a search on the OHLC values, I think it's quite safe from people "recognizing" the data. You could also sector indexes or back-adjusted futures contracts. I'm sure even if you picked a market I've traded before I wouldn't recognize the pattern unless it was very recent. It's probably true for most other traders too. You could also do other things like... inverting a chart. That should prevent traders from "recognizing it" but the chart itself would still have the same TA characteristics. I like the idea of using non-US stocks too.