I can't believe the shit these kids today call an "edge". An edge use to involve arbitrage and an inefficiency. roughly speaking, 80% of the time it chops and you will make small money adding to losers. 20% it trends and you make big money adding to winners. When it's all said and done, you break even minus the spread and commissions. But if you have courage, patience and flexibility (the three qualities neccessary to all success in any field) you will be able to identify inefficiencies in the current strategy you are involved in and act on it to your profit.
I think that just using the tape should be sufficient. If you are reliant upon any visuals or indicators whatsoever, I think you put yourself at risk unnecessarily. There are subtleties of price action that don't show up on charts and indicators, especially as regards volatility. The interesting thing about the "variants" is that the further along in the use of the algorithm you get, the more variants you identify. You could almost make it an axiom of system development, I think. I'm not sure if that process continues forever, but it certainly endures beyond the stage of going live, even, since not all structural variants will show up in your backtesting. The judgment aspect of it is in figuring out which are variants which will need to be filtered and which are going to be tradeable. As long as you can keep identifying objectively different scenarios, you can develop a database of historical outcomes for them. What do you think of a metric of time in between trades as a measure of sensitivity to noise? Might it be the case that the greater the range of time in between trades to the minimum possible time between trades tells you something about the ability of the algo to avoid overtrading? So you could say that a system which could theoretically trade every day (e.g. a system based on an opening range breakout or the relationship of the open to some pivot point from the prior day or something like that) which has a range of time between trades of 0 days (i.e. at least one instance of two consecutive days of trading) and 10 days is less sensitive to noise than one with a range of 0 to 5 days. That seems kind of intuitively valid to me, but what do you think? And Hershey was one of the quacks this thread was aimed at, actually. I saw a repost of some post of his (he's on ignore, thankfully), where he was talking about some 81 point cycle or something and I was like "The market doesn't give a shit about the number 81 at all". If you've got the number "81" hard-coded in your model somewhere, that model is a POS.
I only want to be in the market when it's trending on my timeframe. Once it stops trending, I just wait until the next trend looks like it is starting. I'm not interested in trading the chop part. It is true that there is a great deal of non-trending time in the market, though. Much of my day is spent updating spreadsheets as markets move and changing the prices on my stop limit orders in response to what I see happening. The actual amount of time spent in trades is much smaller than the time spent doing those things. I prefer that to the frantic jumping in and out some people do, though.
Jack, I think you are about to regain all your credibility on ET ... just post your account trade log for today and enjoy!
what makes you think you are so special? I also don't want to be in the market when it isn't trending my way. But you know sometimes you have to pay the bills, and I never know when I get in if this is somehow going to be the time it trends my way. That putting it on and getting stopped out will overtime eat you alive. But it would be nice if it was that simple.
Seriously? And looking at the gravel on the sides of the road when you are driving I suppose I was checking TickData's historical tick data quality just this morning, comparing with my own data collected through IQfeed historical server ... even with the proper tools, it was a major pain. And I wasn't trying to read the market, just looking at highlighted differences. Anyway, regardless of the type of timeframe and settings used for it, going from tape to bars renders a huge service, even though it is in all cases very improperly done ... that service is filtering. Not to mention that the brain can easily identify & remember the visual aspects of patterns, try that with the tape. I honestly believe this isn't relevant. Take a HFT system w/ P/F 1.5, randomly cut 90% of its trades, the result is a no-longer-HFT system which P/F remains 1.5. Of course, it is just as sensitive to noise as the original one. In my books, noise can only be defined in relation to a given type of signal ... when the signal characteristics don't overlap the noise characteristics, all is good and easy ... but when the noise characteristics overlap the signal characteristics, it becomes difficult to impossible to differentiate signal from noise at 100% sucess-rate; as a result, you can either filter a lot, and lose some of the signal, or filter a little, and interpret some of the noise as valid signal. Again, I believe most people don't take any type to define & refine their signal characteristics ... so of course, they experience a lot of overlap with noise.
Noise is easy to define. Noise is a time frame where your trade setups no longer work. For me 1 min is noise. For some 1 second is noise. For other 1 nanosecond is noise.
zero hard-coded parameters: http://upload.wikimedia.org/wikipedia/en/math/b/f/2/bf2d70f3dd440e66c83ea90c54f8b9ed.png The SE is therefore a new concept in itself; as Feynman put it: âWhere did we get that (equation) from?. Nowhere. It is not possible to derive it from anything you know. It came out of the mind of Schrödinger.ââRichard Feynman