Hi all, I believe the significance of Nuke's method is that it uses a Bayesian conditional probability in predicting whether a bar is going up or going down, based on the 90th percentile LSD SLD: Let's say you are betting long. If the bar has gone up beyond open+LSD90 then the probability that this bar is an up bar is 90%. That's because you have surpassed 90% of the times that the bar is a downward bar (i.e., most downward bars will have the short distance going upward). So then once it surpassed open+LSD90 (your limit buy price), you can then impose the SLD90 as your take profit point. That point is guaranteed to be reached 90% of the time IF you are in an up bar. So the overall rough probability of being in profit is 90% x 90% = about 81%. Of course there is a chance that once you have surpassed the LSD90, it could swing back down to the open and in fact it can go down to the open - LSD90. That will probably screw up the stats. I believe this is where Nuke goes into tick data to find out how often that happens and perhaps adjusts the limit price upward to reduce that probability. I can see however that you still may retain some statistical edge even with this momentary swing down probability because it is likely rare (I need to check) that such an extreme swing down occurs once you've hit the price open+LSD90. The trouble comes in when LSD > SLD. That means that the distribution of SDs overlaps the distribution of LDs, which means that you won't be able to discriminate up or down bars based on the LSD or SLD. This is where I have lost Nuke's logic. He says you can go ahead and move over the SDs that are greater than the SLD to the LD column. I think he explained it once but I still don't understand how that changes the probabilities. He says something like if the bar is so volatile such that the LSD is > than the SLD, then you could have bet either way and made a profit or loss. I believe he might have some idea in here, but I am still not following that logic. If anyone can explain this logic, I would greatly appreciate it.
hi kidPWR I have been applying it to stocks as well and I am finding that the LSD-SLD spread is just not enough to cover the commission and slippage when I use it on daily bars. I haven't surveyed all the stocks to see if such a spread does exist consistently for some stocks. I agree that if one can apply huge leverage to that LSD-SLD difference, then it is likely a viable strategy.
Hi Nuke, This is an awesome post and I want to thank you for opening my mind to this idea. I've used your ideas for a long time and I'm happy to share some ideas. - I continually calculate the Expectancy for each trade. ie. if I'm making 2pips and losing losing 9pips at 80% hit rate then even if 80% hit rate holds I'm going to lose money in the long run. So I don't trade when the Expectancy is < 0. - I discovered that failed moves to the target don't only occur in the opposite direction. Price may trade up 4pips, revert to the open, trade down 9pips, revert to the open and then trade 20pips lower. Testing on hourly bars would show a profitable trade when in reality the trade would have been stopped out before becominig profitable. Ideas and comments welcome (just don't tear me apart)
Just discovered this thread - good ideas to pursue here. But curious if anyone is successfully trading these ideas - because there have been conflicting backtests presented - some showing profits and some losses (when considering tick data, which would be the case in reality). So, curious if anyone is actually trading the method as presented "successfully" and consistently overcoming spread and commissions. ?
Frankly I never got it to work though I pick it up and toy with it occasionally. The basic concept is to use a basic filter to give an edge (EMA 20). Then look at the most recent stats for various PA movements and make a trade on the most profitable settings. It's like backtesting in real time, always optimizing for the latest working settings. Here's the basic question: What percent of the time did price move x pips from some starting point in a favorable direction without moving more than y pips against you? You find percentages over ranges of values for x and y and pick the "best" one. I'd evaluate "best" as "highest expectancy". OP's initial method of deriving this info from OHLC bars is actually pretty complicated. Doing this with tick data is more straightforward. That's it in a nutshell.
I didn't say it right above and can't edit the post any more. Here's the nutshell concept: "If price moves X pips from some starting point A to a point B, then it will continue Y pips to point C without returning to point A with a probability P."
We managed to program this using the "tick" version he described for more accurate data. I'll state the issues I encountered with someone whom I worked on this with on equities: 1. It is profitable not including commission, with a PF from 1.5 - 2.5 on most issues. 2. As soon as commissions are added (at least on a retail scale), the system was barely profitable and sometimes negative. Also, sometimes the amount of leverage the system was trading was pretty crazy as it grabs only a small amount of what the market offers. The solution could be to trade prop or trade other instruments with lower commission for the leverage. I did not find it had a large enough edge to drop what I was doing, but in the future, it may be part of what I do. Also, I never had accuracy that approached 80-90%, I was in the realm of 60-70% even with the tick method.
nope im still around, just busy.. the problem is that people want a solution. while I was showing how to look at the market and apply % for people to evaluate setups based on price formations.. for example people take the tick data thing to the extreme while i was just showing one way to find an edge... the goal of this thread is to enhance what your already doing with statistics... for example if your trading .... AUD/USD you can say on the weekly im against support after resistance was broken so i should be coming off support... - this is the bases for your trade you then evaluate the entry to find the most statistical advantageous position.... (the weekly bar coming off support and made a low came to open and thrusted up 20 points I know that most candle who move 20 points continue to move another 70 points once they do this 90% of the time) The interest rates provide trickle into the direction of the trade and I know that X bars form in direction of interest rates increasing the probability of this bar to be in direction of possible trade... The idea here is to gain entrance profitably in direction of where you think the market will move while limiting possible losses and skewing your odds you enter a trade and note the statistical entrance advantage and protect yourself creating large right tailed profitability bell curves.. this guarantees that you will be profitable over the long term...