looking at random charts will show that price ONLY isn't enough. -Add volume and you get something usefull -Add fundamentals and money management and you might make money every single year.
Depends on what methodology you are using. Assuming the underlying meets minimum threshold of liquidity, many traders have done very well just trading currencies looking only at price. Using random charts can indeed be useful to test methodologies that require a distinctly non-random characteristic, such as persistence of trend over time.
Have you thought about having a chimpanzee program the excel macro? That would really make it random.
Nice thread. One thing it has convinced me to do, is to generate several sets of random data and run it through my back testing. Then compare the charts and graphs I pump out of my historical results of real data vs random data. I will expect a loss with the random data, due to slippage and commission costs. I will expect wildly different results. I will expect parameter changes to cause erratic changes in profit, whereas profit changes slowly with parameter changes in a robust system. This will be yet another curve fitting detection method. Should also help catch programming accidental edges into a system. ( Which I do frequently ) If backtesting against random data consistently produces profit, then you KNOW you have a bug in your algorithm somewhere. Nice...very nice.... peace axeman
axeman, The key to doing the types of random data testing you are thinking about is to do lots of them. You need a significant sample size to draw any useful conclusions. Many traders pick a preferred method of trading (breakout, trend following, momentum, contrarian, scalp, whatever) and utilize filters and try to fit the market to their system. Other traders will adaptive will try to fit their system to the market by asking the question: is the market currently trending or in trading range? Breakout, trend-following, LBR grail trades, etc. work best with trends; while oscillator/cyclic methods work best during in trading ranges; etc. These traders modify their methods (whether system or discretionary) to market conditions. Now why not take that one step further? Instead of just asking "is the market trending or in trading range" why not also ask "does current price action show behavior closer to random or persistence". Based on the answer to this question you can shift your trading strategies. And I don't just mean entry-and-exit signals, but how about position sizing/risk? When markets show higher propensity for random fluctuations, either don't trade at all or trade less size. When price action shows clear non-random behavior the odds are much higher that your edge can exist. The accuracy of measuring the degree of randomness is the tricky part. You have a measurement problem in that the more accurate your measurement the more time has elapsed and the more likely that the underlying behavior has shifted. This is the nature of the problem/challenge. I'll conclude by saying a small edge is all you need!
Dottom, Thanks for another thought provoking post. I guess this would be an example of the first category you've mentioned: Instead of imposing a view that one or another strategy is trending vs. reversal vs. breakout vs. etc., I've found it's normally helpful to let actual trade performance data tell me in which environments one or another event is followed by continuation or reversal, and then for future trades simply impose a filter accordingly. Interestingly, sometimes my presuppositions about which environments would be most supportive have been 180 degrees wrong. But I guess this is in part what makes the learning side of systems development so fun. Your final paragraphs, by the way, I find really intriguing. So now I guess the trick for me will be to learn how to measure random-like price behavior. Thanks again for your post. NorskTrader
these discussions sharely get into the practical uses unfortunately. so its random or chaotic or whatever. and randomness can be measured, and autocorrelation can be measured. ok, great. where do we enter and where do we exit.
Traderkay, Try this experiment. Take whatever measure of random vs. non-random price behavior you want, such as Hurst exponent as basic starting point. Say you trade on 15m or 30m bars. Now go to a lower time frame to measure if data is random or not, say 1m or better tick data where you'll see more noise but have higher resolution. Don't enter trades when the tick data analysis shows price activity is closer to random. The sample size here is important. Too large a sample window gives accurate data of what happened in the past, but might be too late for anticipating what might happen in the very near future. Now plot the Hurst exponent like any other oscillator and see if you can detect any patterns? See any empirical correlations between periods of "chop" and "drift" and Hurst exponent? What happens to the Hurst exponent when price makes sustained trend move? What happens right before the big trend move? What about LBR grail type setups- but instead of buying on any pullback to the MA after trend move, what if you use measure of random vs. non-random by analyzing the tick data? What about various cycle methods? Ever try to filter Ehler's cycle vs. trend mode by adding a filter for random vs. non-random price behavior? Hurst is just the starting point but is what most people are familiar with. The small sample size makes Hurst exponent generally inaccurate for most purposes, but you can look at extreme values and get some useful information. Not your basic TA, but definitely the high-tech stuff works. You also have to understand that you're doing a lot of work to uncover just a small probability edge. But that's all you need.