backtesting.py icon indicating copy to clipboard operation
backtesting.py copied to clipboard

Commission fees differ significantly.

Open MrDenfish opened this issue 2 years ago • 0 comments

Thanks for posting this problem, I am having a similar issue when comparing backtesting.py with TV , commission fees differ significantly. After considerable digging I have a theory that backtesting.py is not computing fees on the Exit Price (at least the math points in that direction). This may be my settings or something deeper. Hopefully @kernc will comment...

I have attached two screen shots one of trade results for TV and the other stats _trades. backtesting.py & TV are using coinbases' stupid high fee of .004% no slippage and the data uploaded in backtesting.py is the same data downloaded from TV chart using 1m time frame.

self.buy(size=1500, sl=(1 - self.stoploss) * self.data.Close, tp=(1+self.takeprofit)*self.data.Close)

bt = Backtest(data, scalp, cash=500, commission=.004,trade_on_close=True, exclusive_orders=True)

My observations: TV Entry Price is the same as close and reflects no fee. .2166 BT.py Entry Price shows the purchase with fee applied .2166* 1.004 = .217466 which is correct. TV Exit Price shows .2183 BT.py Exit Price is also .2183 TV shows a profit on this trade of -.06 while BT.py shows a profit of 1.2504

TV math:

  • Entry .21660*1500 = 324.9 *.004 = 1.296
  • Exit .21830*1500 = 327.45 * .004 = 1.3098
  • fees 2.6058
  • profit 327.45-324.9-2.6058 = -0.558 or -0.06$

BT.py math:

  • Entry .217466 / 1.004 = 0.216599601593625 = close (.2166) - looks good! - cost .217466*1500 = 326.199
  • Exit .2183 / 1.004 = 0.217430278884462 <> close (.2183) proceeds .2183*1500 = 327.45
  • fees 1.299(entry) + 0 (exit)
  • profit proceeds-costs or 327.45 - 326.199 = 1.25 - math checks out -

In a nutshell TV is computing fees (entry price * size)* 1.004 where as backtesting.py appears to be computing commision by (Entry price * 1.004)* size. I think both methods are correct, the results are the same. However, TV computes both purchase and sale, BT appears to compute on the purchase only.

backtesting py TV

Originally posted by @MrDenfish in https://github.com/kernc/backtesting.py/discussions/734#discussioncomment-6200892

MrDenfish avatar Jun 17 '23 16:06 MrDenfish