Here's an example of a bar-by-bar analysis in which I take a trade immediately upon close of a valid signal bar, with initial stops left in place, never moved to break even: Sh @ 76.44, sl @ 76.50, pt @ 76.14: pt filled for +$300 Sh @ 76.27, sl @ 76.42, pt @ 76.99: pt filled for +$280 (if able to move pt, then more) Lg @ 75.72, sl @ 75.63, pt @ 75.94: pt filled for + $220 Sh @ 75.75, sl @ 75.95, pt @ 75.46: pt filled for +$290 Lg @ 75.42, sl @ 75.30, pt @ 75.78: pt filled for +$360 (if able to move pt, then more) Sh @ 75.79, sl @ 75.87, pt @ 75.55: pt filled for +$240 Sh @ 75.60, sl @ 75.70, pt @ 75.40: stopped out for -$100 Sh @ 75.67, sl @ 75.81, pt @ 75.49: pt filled for +$180 Lg @ 75.35, sl @ 75.24, pt @ 75.58: pt filled for +$230 Sh @ 75.81, sl @ 75.94, pt @ 75.57: stopped out for -$130 Sh @ 75.83, sl @ 75.96, pt @ 75.57: pt filled for +$260
Would you mind explaining how you perform your bar-by-bar analysis? Print out your chart and do it by hand or scroll back to the start of the day and advance chart 1 bar at a time? thanks
Was thinking the same (I did find backtesting by hand useful at the beginning, during the initial stages of learning, etc., but after a while, IMHO it makes more sense to run proper, automated backtests - unless you're willing to manually backtest at least several months of data at a time, that is :eek: ).
Agreed. Backtesting by computer has the added advantage that you can run many what-if's a day to refine the strategy or to take a quick look at the brain farts all of us have while watching the market.
As CRG Trader pointed out and you yourself acknowledged, the discrepancy between your theoretical/paper results and your actual results lies simply in that when you backtest "without cheating" you only look at the price action of the instrument in question, whereas in real time you're being conditioned by other developments in and outside of your market. This is actually good news, because if the reason for the discrepancy was other (inadequate backtesting, a seemingly unbridgeable psychological gap between trading paper and real money, etc.) you wouldn't have any obvious, clear-cut, immediately implementable solution, but... If I may ask: how can fully automating a bar-by-bar trading system take more time/effort than mastering your own [psychology/human tendencies when applying your] trading method? I don't see how automating a specific trading system based on pure price action in any given instrument could take more than a few hours? Which brings us to... OK, so last Friday you would have netted 2,130 USD with one car - and after I recently finished reading your whole "oeuvre" (thanks for sharing, Nod - and +2 for joining the rock band! ) I'd say it's fair to assume that you could probably trade safely with 2-5 cars. If you backtest your strategy (not just for a few selected days and, as I'm sure you already know, including commissions and realistic slippage) and you actually have an edge half that good with a drawdown that you find reasonably easy to stomach, why not simply trade it mechanically? I didn't get from your posts what platform you're using (other than IB), if you're using NinjaTrader and the holdup is technical/programming I'd be happy to help (as I'm sure some other people would), or simply cough up a couple of hundred and hire a programmer. If it's an ego thing (like wanting to master yourself in an epic quest rather than relying on an cold and dull algorithm), well, I can sympathize, but for what is worth, if I had an edge half that good, I'd milk the f#@~ out of it (and hire someone later to boost my ego ) And I still think you're a whiner
Perhaps you should re-think #1. If there is no bias, and the valid set-ups are taken randomly. The % of wining trades should not be affected by how many of the setups you trade, though you would average more won or lost by taking more trades. Perhaps there is a subconscious bias, or is my thinking wrong?
I scroll back to 7am ET (an hour before I'm awake enough to actually trade) and advance 1 bar at a time. The example I posted was from 6/29. I used to include the setup, the signal bar, and the reason for the particular entry, stop and target, but after doing that for months, I don't need to do that any more. I only very recently began evaluating the benefit of leaving the initial stop in play until either it or my target is hit, (vs moving my stop to break even after a 10-15 tick move in my favor). Somehow the fact that Geez has been trading that way forever and doubling/tripling his trading account each year was not enough proof for me. I had to actually try it in my backtesting, and have only been doing the bar-by-bar that way since late June. I'm married to a programmer who has already programmed 3 different ATS strategies and has been testing and tweaking them and always seems within 100 yards of the finish line, but you know how them geeks are, they just can't seem to stop tweaking... We've run thousands of automated back tests, but they don't relate well to reality for some reason. I know my manual backtesting is valid because I not only prove it each day after the market closes, I discuss signals I see in real time and often don't trade, only to watch them unfold wonderfully without me. The bar-by-bar is so easy because I have nothing else distracting me, no other charts, no streaming news, nothing, because I'm focused on only the 5-min CL chart and I can make decisions one after another as I reveal the bars. But in real time a 5-min bar takes 5 minutes to print and while waiting for it I look at other stuff, or take breaks, or find reasons why it's one of my setups, but... Back to automation, the ATS systems my husband programmed are based on some very simple ideas I had well over a year ago. Some worked better than others. I may end up working with him on automating my strategies with CL and giving it a run if I don't get out of my own way soon. I agree that I should comfortably trade 2-5 lots CL and have traded that size several times (but not comfortably at all with 5), so went back to 1 until I can mechanically take all trades off valid signal bars during high probability times of day when the moves tend to be the strongest. So my goal Monday is to absolutely focus on CL and nothing else, because during the bar-by-bar analysis I have no idea what the ES is doing, what Obama/Geithner/Bennie-Boy are saying about the economy, where the stochastics are on my USO chart, etc. I'm looking at the price action prints and nothing else. I'm really beginning to understand why so many traders say to ditch all indicators.
I have frequent bias, conscious and, I'm sure, unconscious. Bias has cost me (and I'm sure most traders) more money than all other mistakes combined, both in terms of large and unnecessary losses as well as potential large profits. Matcha posts "cost of mistakes" each day in her journal. I would say that 70% of my cost of mistakes would be a result of bias.
Thanks for the detailed post. I don't know what charting software you use, I use Sierra Charts and you can enter trades directly from the screen along with targets and stops. (I know Ninja also has this feature) It would seem like a natural compliment to your style of trading, since you are clear when you want to get into a trade (I am attaching a chart that I think is your 1st short on 6-29-2010 to give you an idea of what it would look like) If you are placing your trades on the chart it may help you to stay focused and once you have entered the trade on the chart if you get side tracked at least you should be in if it goes your way. Finally if your focus is on placing the trades on the chart to the best of your ability it might take some pressure away from the actual outcome of the trade.