Forums
-
BEAR BULL TRADERS FORUMS
-
- 229
- posts
-
- 2,958
- posts
-
- 11,596
- posts
-
- 407
- posts
-
- 73
- posts
-
- 173
- posts
-
- By silverrocket
-
Trading Psychology
- Ask Dr. Katz your Psychology Q&A Webinar
- Chat Room Psychology Alerts
- Psychology Resources/Journals
- Psychology Articles
- Averaging Down
- Being Mindful
- Building Mental Resilience
- Coping with Anxiety
- Dealing with FOMO
- Developing a Pre-Market Routine
- Fear of Taking a Trade
- Improving Discipline
- Increasing Patience
- Overtrading
- Other
- 686
- posts
-
- By KHarp83
-
- 418
- posts
-
- 85
- posts
-
- 89
- posts
-
- 73
- posts
-
- 152
- posts
-
-
Topics
-
-
RECENT POSTS
-
1 min ORB algo. Here is the backtest for the 1min NVDA Long that I created (R=$200) I have been mining the data to see the possibility of making ORB algos for the last couple of weeks. I looked at 1min ORB first using NVDA long direction. There seems to be about 2.5 years of data where ORB setup could be used. I did try and read a few algo books, but didn't find any useful. Some interviews in the Market Wizards books I found very useful for algo trading. The rest of my method I figured out on the way. I have been profitably trading algos on Futures for almost 2 years now. The only issue I have is my algos trigger a trade at most once a week. Not enough. So looking for more setups. So I first estimated my parameters and values based on what I have learned from the mons at BBT. I then backtested only the last 3 months using Ninjatrader and see if a profitable algo is even possible. This time it was, but these results are useless except for feasibility. I then walk through each parameter to see what has correlation. Interestingly enough almost all the parameters have value limits but not correlation. Except for volume, which the larger the better, and the distance from VWAP, none of the other parameters had correlation except for limits. Though not expected, this is great because it means the parameters are robust. Though Ninjatrader has an optimizer, it usually is not very useful since it curve fits too much. So I manually optimize then run the parameters for the last two years and got mediocre data. But that's normal. Then I use something I learn from one of the Market Wizards, where he uses 2 years ago for optimization and runs the algo on 1 year old data to see if it works. So I noticed, when I used the 1st algo version we were in the red about a year ago. So I used the red time period to optimize. 1min ORB profit when optimized on the recent 3 months then back tested for 2.5 years. I optimize for robustness not profit. I use to make spider web plots (in Excel) with each strand a different parameter and thus try and create a large stable sweet spot. Where the parameters could move around and it will have little affect on the stats. There is almost always a plateau like that, but sometimes it is under water. So if I let Ninjatrader optimize it gives an outragius profit factor of 5.7 with a highly curved fitted solution that if you move any parameter, even slightly, the profit drops off fast. My optimization has a modest PF of 1.7. But all the parameter correlation are gone (even Volume and VWAP). You can move parameter values around, even large amounts with little effect. All the parameter values become min or max limits only. Thus these are the edges of the optimization plateau. Here are the parameter limits: Volume > 300K 1min candle body > $0.60 1 min candle < $2.7 Top 1min candle wick < $1.10 How far above VWAP when initiating trade > $0.20 and < $2.30 Market opening price Gap for NVDA >$ 0 SPY 1min candle close > -$1.00 Essentially, the algo is saying don't take a NVDA Long when SPY is careening lower. The target and stop are loosely optimized, but are not meaningful since I have a separate trade management algo that takes care of that. For this back test I used a 3 to 1 traget to stop ratio. Then I ran it on the full 2+ years. the data you see is above in the opening graph. Below are the numbers. The max consecutive losers is a little high for my comfort, but its good enough I will start running in sim and create the short version of the algo. Then move on to TSLA, ES and NQ. Thanks for reading to the end. Rob C -
-
-
