Reality Check for this strategy with -99% Z-score
15 replies
fiverr
8 years ago #114481
Hi,
Please let me know your thoughts on this strategy report. How is the sharpe ratio calculated? Is it daily, weekly, monthly? Is the profit factor acceptable? I read some places where 1.5 is good while I read other places pf>1.25 is good enough.
Is this strategy too good to be true? Can someone help poke holes in this report? Constructive feedbacks are greatly appreciated.
tomas262
8 years ago #134216
Its normal to have low PF when it trades often and does many trades. PF is just one piece of a puzzle. But to me it looks promising. Have you done robustness tests already?
fiverr
8 years ago #134218
Tomas,
Thanks for your reply. Can you elaborate on the robustness tests? This strategy has been tested from 2000 to 2015 i.e. 16 years.
mikeyc
8 years ago #134220
I suspect there’s something wrong with the strategy or the setup, i.e it’s not a realisitic. Have you tried backtesting the strategy in Metatrader?
PS. I’ve seen this sort of thing where the spread is somehow set to zero with no commission, so that tiny tick movements are turned into a large number of profitable trades. Put in the correct spread and ECN commission and the profitibility vanishes….
RJL
8 years ago #134221
What Test Precision mode is this strategy tested on? 1 Minute Data, Tick Simulation, or just Selected Timeframe only?
fiverr
8 years ago #134227
I suspect there’s something wrong with the strategy or the setup, i.e it’s not a realisitic. Have you tried backtesting the strategy in Metatrader?
PS. I’ve seen this sort of thing where the spread is somehow set to zero with no commission, so that tiny tick movements are turned into a large number of profitable trades. Put in the correct spread and ECN commission and the profitibility vanishes….
MikeyC,
Excellent question. I am testing the portfolio in MT4. Similar to you, I fell for the zero spread in the past. Currently, I have the spread as 3pips for EURUSD and 4 pips for GBPUSD.
fiverr
8 years ago #134228
The average bars in trade figure looks suspect. It suggests (correct me if I’m wrong) that the trade is opening and exiting on the same bar; if this is the case you should retest it using real tick data with variable spread.
Notch,
Good question. I am testing this as a portfolio; hence, the bar number is not reporting correctly. The average hold time for each trade is 35 hours.
mikeyc
8 years ago #134229
What does the equity curve look like? Are you testing using real tick data or tick simulation using M1 data in MT4?
I don’t get how you have 114K trades with 35 hours average holding time unless there’s more than one trade open at a time? That’s 455 years.
What is the strategy like on it’s own? It’s more likely to be iffy portfolio calculation IMHO….
fiverr
8 years ago #134251
What does the equity curve look like? Are you testing using real tick data or tick simulation using M1 data in MT4?I don’t get how you have 114K trades with 35 hours average holding time unless there’s more than one trade open at a time? That’s 455 years.What is the strategy like on it’s own? It’s more likely to be iffy portfolio calculation IMHO….
Yes, I am testing using M1 data and am holding multiple trades at a time since each strategy is independent of the others. Below is a snapshot of 1 strategy.
mikeyc
8 years ago #134255
Looks good, I would put it on a demo account to see if there’s some weird fluke.
fiverr
8 years ago #134256
First merge all the strategies into one then upload the QA file . It is hard to make comments of value with so little information.
The general aim of building a portfolio is to reduce performance variance. For me, the maximum consecutive loss figure of 42 exceeds my preference. By adding low correlated strategies to a portfolio, I aim to minimise a number of things including consecutive losses. Have you run the portfolio correlation analysis in QA yet?
Notch,
Thanks for sharing your system info. I notice that that your winning percentage is more than 66%; hence, the consecutive losses should be lower compared to a trading system with 46% winning trades. You can try searching on youtube/google for monte carlo simulation. You will find that the maximum consecutive losses is dependent on the winning percentage and not the profitability of the system. You should consider reducing your drawdown as your return to drawdown ratio is only 11:1 i.e. you are risking $1 for $11 gain. A good long term system should be more than 25:1 i.e. $1 risk for $25 gain i.e. better odd.
fiverr
8 years ago #134257
What Test Precision mode is this strategy tested on? 1 Minute Data, Tick Simulation, or just Selected Timeframe only?
Sorry for missing your post. I am using 1m data for testing.
mikeyc
8 years ago #134261
Modelling quality of 25% !!!
https://www.mql5.com/en/articles/1513
As we mentioned before, there are just 4 reference points for a one-minute bar to be formed! This is why the rating of modelling quality for one-minute bars does not exceed 25%.
fiverr
8 years ago #134262
Modelling quality of 25% !!!
When you test on 1M data, the modelling quality will be 25%. If you test on the trading time frame, you will get 90%.
mikeyc
8 years ago #134265
I only test using Birt’s Tick Data Suite 99% backtesting quality using Dukascopy Tick Data as a source. I was merely pointing out that 25% quality is the best you can get with MetaTrader without external tools and real tick data.
fiverr
8 years ago #134268
Yes and? You either build a tick app or you buy one. Buddy, I don’t know about you but nothing less than 99% will do for me. Mikeyc, don’t tell me you’ve been testing scalping systems using 1min data!
Actually, I am quite familiar with Birt’s tick. If you design your EA correctly, you will save yourself lots of time without using tickdata. You may use it for final testing but not during development. If your EA is not coded properly, a slight variation in price feed from different brokers will cause your EA to fail in real time. You should go to MQL4 market and search for the EA called Abbey 24EAs portfolio. You can test this EA with open, control or tick price and you will obtain similar results +/-5% in profit and drawdown. Someone else also tested the same EA using Birt’s tick, it took them many hours to run the test. At the end, they got similar profit compared to MT4 1M open price test. Your testing methodology is dependent on your preference. Do you want to get the same results in 30 seconds or 6 hours? There is no right or wrong way of testing an EA. If you coded your EA correctly then you can save yourself many hours without waiting for the tick testing to complete.
Viewing 15 replies - 1 through 15 (of 15 total)