You may also find, however, that your "good ATR-based system" isn't as good as you thought it was and you'll be faced with the prospect of starting over. Don't be disappointed if this is the case. It happens. Often. But the re-working and re-working is where winning traders come from (losing traders won't bother, or will give up).

After live trading our backtested systems for three months, I am pleased to see it has reflected the backtested results. However the backtesting must be as DbPhoenix describes - there are no short cuts and detailed testing must be done with win/loss ratios, profit and loss ratios, and strict risk management rules; it is a lot of work; constant review to improve and then re-test again. Correctly backtesting your strategies, following all of the rules, all of time, following your trading plan - no exceptions is really the holy grail of success. You can say with absolute certainty that going forward, the markets will never perform exactly as they did in the past and that live trading isn't exactly like testing, etc. But for the most part, this is the exception rather than the rule and should not be used as an excuse not to test.

The goal for any back test should be 90% or better accuracy in simulated trades vs live fills. This can be done by using accurate commissions, slippage, and understanding the limitations of your back test engine. Being overly pessimistic or overly optimistic is a mistake, you should try to be as accurate in assumptions of the market as possible. But in general it is safer to be overly pessimistic.

Back testing and modeling of trading strategies comes down to 2 essential types of assumptions 1) assumptions about the markets and its micro structure 2) assumptions about the type of activity that is exploitable to generate alpha. You must make both assumptions as close as you possibly can to real life as possible. The more off you are in your base assumptions the more off your back test will be, and that is completely human/user error. No institutional grade software, back tester can help you fix this.

In regards to strategy type affecting back tests, this is just a falsehood. It is not the strategy that is affecting it, it is the poor back tester. Most people are using tradestation, ninja trader etc. These are not very robust back testers and were not built to handle intraday or scalper type strategies. They are based around the idea of re hashing common technical indicators and using longer time frames. Just you need to choose the right tool for the job. The backtesting engines in Tradestation or Ninja are just not the right tool.

What do you recommend as a solution? Thank you.

Well the best solution would be to have a fully extensible back tester that can be customized to fit your strategy. However the only way to do that would be to either program one yourself, extend one from a package (R quantstrat/ Python bt) , or buy an institutional back tester. There are no retail backtesting platforms that are fully extensible to my knowledge, that would include trade station, TOS, ninja trader, multi charts. I have gotten multi charts to about 85% fill accuracy, but it ment that I was limited to types of strategies it would do well with. Long term daily bars and using market orders. Also I had recorded the liquidity patterns in order to estimate my slippages accurately.

However here are a few tips that can help:

1) if using limit orders use 1 tick pass through for a fill, not on touch.

2) record the level 2 and determine the average bid ask spread, as well as average sizes on bid offer. This will give you better slippage and fill assumptions.

3) understand the limitations of your back tester. Try to test systems that are within the boundaries of what your tester is good at, or understand where it can be giving you over or under stated results.

4) learn how or when your live system actually runs the logic cycle. Is it at end of bar, each tick, what is the order of logic priorities.

5) Use a broker server live sim and small sized real money to compare results. Broker sim will only show gross errors as it is still a simulation. You need to use real money to find out if your slippage, latency, and executions will be in line with what you expected. Then go back and add these new adjusted numbers to your test to find more realistic result. THIS DOES NOT MEAN GO BACK AND CHANGE RULES! it only means go back and see if you were over or under estimating the results, and see what it would be like now that you have experienced real fills.

That's great information. By 85% fill accuracy, how inaccurate are the remaining 15% of the orders?

I am at the point in my trading where I need to backtest and the learning curve for backtesting just as insane if not more insane than starting to learn about all of the concepts of trading.

I have created a thread:

Quantatative Trading: Intraday Options Data and Analysis Software I am at a point in my trading journey where I want to perform backtests with the intention of implementing an algorithmic trading strategy. I have found a source that has the data, TickData, and software to run the backtests, Multicharts. I have …

. I am looking to trade options intraday with the lowest timeframe being 1 minute bars. I also want to develop other strategies, but that's probably the most demanding of them.

Do you have any recommendations for my situation? Thank you.

In regards to how the 15% of orders were. they were scattered. But about 2-4 x my estimate on slippage, i did have some really hurt though 10 x+.


I suggest you learn to make algos on any single time series like futures, stocks or FX before dealing with options based strats. They are much more complex and require a hell of a lot more software/hardware. I have a thread i started here dealing with that, basically no retail package fitted my needs. So we had to go enterprise level and build out from there.

Options Backtester for Algorithmic Trading I have been trying to find a solution for backtesting Options strategies? I have not seen any packages readily available. Is there any software out there that does it? Or would i end up needing to write a custom solution?

Thank you again.

What timeframe were you working with? I am curious if you were using really small timeframes. The smallest I would want to work with are 1 minute intervals.

Daily and 4 hour. I dont trust the execution for anything smaller then that on MC or tradestation ninja etc.