where does slippage come from

[t3chap]

For backtesting, where does pysystemtrade come up with slippage values? Are these assignable/editable?

I want to see how sensitive my system is to changes in slippage but it isn't obvious to me how to do this.

Alternatively, can you set a specific value for commission? That could also use to determine the effects of transaction costs.

Can this be set using yaml file? Set individually or for all of the instruments?

Thanks for any help.

[tgibson11]

Spread costs are stored in the MongoDB spread_costs collection. They can be imported from this file, using this script.

Commissions are here, in the PerBlock column. They are read directly from the file, so just update the file and you're done.

Additional info:

• There is a feature in interactive_controls to "auto" update spread costs using a combination of sampled bid-ask data (collected by the stack handler), bid-ask data from trades, and configured data. It does allow manually setting values, but if I recall correctly, it doesn't have an option to only update specific instruments - you have to go through all of them, so not very convenient for targeted updates.
• There are columns for percentage-based and per-trade commissions, presumably for use with brokers that charge commissions differently than IB. I would not blindly assume that those values would be used correctly (or at all).
• It would probably make more sense to store the commissions in the IB-specific config file. Edit: even though commission data is broker-specific, it is also needed for backtesting which probably isn't supposed to be using the broker-specific code...hmm...

[robcarver17]

Percentage commissions are indeed used, and where they exist in the commissions file they match those charged by IB (just for Korean instruments as it happens).

Yes, if commissions were broker specific you would need to select a broker every time you ran a backtest, perhaps by having the default broker as a configuration option. Personally I wouldn't use this functionality but maybe other people would. It might be useful to have multiple commission files for the use case discussed elsewhere in the discussion.

[t3chap]

Thank you.

Follow up question-
Is there an easy way to put a "penalty" on all of the trades of a backtest to see how much the results change from increases or decreases in commission/slippage?
Something like adding a $5 on top of the programmed commission to verify the robustness of the system?
Really something to account for Murphy's Law when trading something based on estimates.

Thanks for your time and expertise.

[tgibson11]

Depends on what you consider easy.

If I wanted to do that, my first thought would be through the config files: use Excel or another spreadsheet program to derive new commission / spread costs with a formula - no need to edit each one manually.

I'd probably work with the spread costs, since these are 1) far more likely to vary and 2) probably a larger share of trading costs than commissions.