ntw31, your statistical approach is impressive. The approach is the thing of value in your thread, not just the specific system you explained at the beginning. I also liked the pdf you posted on another thread about the study of S&R levels (still reading it). Regarding stats software : I highly recommend R, a free software package. Sent you a long mail about why R kicks ass. Hope you find it useful.
I've been using Scilab (http://www.scilab.org a free and much lighter weight Matlab clone) for a number of years to run my black box (both backtesting and live modules). I only had a cursory look at Scipy/Numpy last year, looked interesting but not enough to sway me to switch since Scilab is extremely code-efficient, library-rich and fast. My system is based on EOD data feeds, may be a totally different criteria with real-time data feeds. D.
Thanks for your educational thread. Is it possible that you explain how you calculated these S & R lines ? I read in a post further that you posted a pdf in another thread about this subject. Would you mind to post the pdf also in this thread ?
nuke, fantastic thread! thanks so much for the work and time you put into it. you're really helping me learn a lot and your approach is understandable and erything. here's a snapshot of my excel sheet for the eur/usd bars from 20090125 @ 23:00:00 to 20090126 @ 20:00:00. hope this is headin' in the rite directions. i can see rite now how in the future it might take a lot of time and i'll be error prone tryna figure out which bars to remove to leave the biggest pip spread between the lsd and sld. is there a formula or something in excel that can do that? the mdrr says i'm risking four and two thirds to make one, while the NPOT says I made an expected 2.4 pips over the last 21 bars. is this about rite? i can send you the spreadhseet to see the math. also, could you tell us which version of statistica you using for your tick formation analysis? if i have this part down then i'm bout ready to go to the next step. thank you very much for takin' the time out to help a fellow southerner ('cept on full moons).
...there are too many bars. ema period + 1 = 21. for starters, you may want to check out the trial version of statistica. nuke made reference to it in his video. the videos are highly recommended.
Armchairtrader, good idea puting the initial approach into steps. I took your template almost exactly and added to it by copying and pasting directly from ntw31's posts . Some of the steps I've clarified with my own interpretation. Hope this helps. Suggestions and comments are welcome. Let me know if you have any questions or if I've overstepped my bounds. This is just kind of what my notes look like. Thx. NTW31 Initial Approach Determine your bias. In this example the 20 period EMA provides the bias. Note whether the bars are above or below the EMA. If they are above, the bias is long. If they are below, the bias is short. Determine the number of bars in your sample. When using a moving average, the number of bars in the sample = EMA period + 1. So, 20 +1 = 21. Determine your time frame or fractal. Weâll use a 3 hour TF in this example. Look at a 3 hour candle's open, high and low. Measure the distance from the OPEN to the HIGH (H-L) Measure the distance from the OPEN to the LOW (O-L) Compare the size of each distance measurement (one will be greater than or less than the other). I donât know what to do if the two distances are equal. Take the shorter of the two measures and put that number in the SD (short distance) cell Take the longer of the two measures and put that number in the LD (long distance ) cell We now have a: 1. SD (short distance column) 2. LD (long distance column) Determine the SD and LD for the last 21 bars in your sample. Take any distance in the SD (short distance column) that is greater than or equal to any distance in the LD (long distance column) and move that number over to the LD (long distance column) In the SD (short distance column) take the largest number over the last 21 bars. This is your LSD (longest short distance). In the LD (Long distance column) take the smallest number of the last 21 bars. This is your SLD (shortest long distance). Determine your accuracy level. It is 80% in this example. Multiply your accuracy level by the number of bars in your sample to determine the success number. 21 candles*.8 = 16.8 candles. The failure number can be determined by subtracting the success number from the total number in the sample. 21 candles â (21 candles*.8) = 4.2 candles. The failure number provides the number of outliers to remove from the sample. Remove the four values (rounded failure number) from the sample that creates the biggest difference between the LSD and SLD. For each value just removed, also remove the corresponding long/short distance for the same bar. The whole bars are removed because they are seen as failures. Measured from the open of the next bar and in the direction of the bias, the LSD + 1 marks the entrance level. Measured from the open of the next bar and in the direction of the bias, the SLD -1 marks the take profit level. The stop loss level = entrance level - (LSD + 1). This is the open. Risk = entrance level minus stop loss level = LSD + 1 Reward = take profit level â entry level Market Defined Risk Reward = Risk:Reward. Market Defined Risk Reward (MDRR) gives you a ratio of how many pips you stand to lose compared to how many pips you stand to make. Based on the stats, this is the risk:reward the market says itâs offering you to attempt to trade at your accuracy level. NPOT is the NET PROFIT OVER TIME. NPOT is simply NET PROFIT OVER TIME or how many expected profitable trades * take profit distance from entrance - how many expected loss trades * stop distance from open This will give you Net profit you would have expected if you had traded SLD and LSD at your accuracy level over N amount of bars. The entrance and exit of a trade should always be in the same bar. If the entrance condition is satisfied in the late phase of the bar formation then the decision of whether or not to enter can depend on just how late the signal was generated. If, for example, an entrance signal is given on a 3 hr bar that has to run 32.5 pips in 1 min then itâs probably a bad idea to enter. Additional Notes: The LSD and SLD can give clues that the market has been overly volatile or not as volatile and doesnât fit with statistically normal behavior. When these clues are recognized the trader should wait for the market to straighten itself out before trying to enter. For instance, if the MDRR is 5:0 so that for every 5 you risk you get 0 in return, that is not a good risk/reward and the market tells you to stay out. If the NPOT is negative, this also tells the trader to stay out. The accuracy level filters out the extremes creating risk and, by association, reward. The âextremesâ are usually in the top 10-15%. The accuracy level allows one to create inefficiency in the machine because a perfectly efficient machine would allow only for the following: you take zero trades, you enter and lose on the trade, or you enter a trade and win but get eaten by the spread. With this initial approach we do not know exactly how the bar develops because we canât see exactly how it ticked up and down before close. Therefore, our stats are based on incomplete price history and we must allow for about a 20% statistical error. The true accuracy isnât 80% or 90% (depending on the accuracy level you initially choose). The true accuracy is the intersection of your bar formation % after the PNR (point of no return) * 1-statistical error. If you choose 90% to filter extremes then the true accuracy is 90%*80%ish or about 72%ish. This can still be tradable, itâs just not as accurate as it might originally seem. Remember that the NTW31 initial approach is a simplified method for calculation of statistical bar formation which actually requires a much more rigorous and time intensive process involving a stats text book, a TI-86 graphing calc and lots of paper and testingâ¦or a stats software program. The next step in the approach is to apply the same logic to tick data so that you know your direct odds. NTW31 posted on the next steps. It is very important to read NTW31âs early posts on trend vs. bias, emaâs, S&R levels, profit extraction vs. prediction, etc. This is the foundational material which leads up to the beginnerâs approach to direct statistical trading.