Invalid or erroneous drawdown calculations in equity plot for resampled datasets

[eervin123]

Expected Behavior

Plots, where data was resampled due to >10,000 bars, should calculate max drawdowns and drawdown periods similar to those computed in the _compute_stats method

Actual Behavior

It appears when plots are downsampled in the Backtest.plot(resample=True) the drawdown calculations and possibly drawdown duration (beginning and ending periods) are miscalculated.

Here are two examples of such charts:
https://www.realitysharesadvisors.com/fund_dash/fix_this_eth.html
https://www.realitysharesadvisors.com/fund_dash/fix_this.html

Note the legend in one of the charts showing 0 days of max drawdown

Here is an example where the red line is misplaced on an incorrect drawdown.

Steps to Reproduce

1. I was using 5-minute candles on multiple years of data.
2. Run Backtest.plot() on a large dataset and it should reproduce the result...Hopefully, if not, I am happy to privately share my datafiles and strategy class for you to reproduce on your system.

Additional info

This was also referenced in a different issue #156 (comment)

• Backtesting version: 0.2.3
• bokeh 2.2.3
• python 3.8.5
• pandas 1.1.3

[kernc]

Plots, where data was resampled due to >10,000 bars, should calculate max drawdowns and drawdown periods similar to those computed in the _compute_stats method

Resampling of equity curve is computed this way:

backtesting.py/backtesting/_plotting.py

Line 118 in 05329de

equity_data = equity_data.resample(freq, label='right').agg(_EQUITY_AGG).dropna(how='all')

backtesting.py/backtesting/lib.py

Lines 62 to 66 in 05329de

	_EQUITY_AGG = {
	'Equity': 'mean',
	'DrawdownPct': 'max',
	'DrawdownDuration': 'max',
	}

('Equity' should maybe aggregate with 'last'. 🤔)

---

Interestingly, the whole thing works correctly (conjecture) on this toy example:

from backtesting.test._test import SmaCross
bt = Backtest(GOOG, SmaCross)
bt.run(fast=3, slow=8)
bt.plot()

Resampling daily to bi-monthly:

bt.plot(resample='2M', superimpose=False)

Max DD Duration seems to be rounded to round units (10 2-months).

Therefore, it must be something specific to your dataset and something in the way drawdown intersections are computed:

backtesting.py/backtesting/_plotting.py

Lines 277 to 290 in 05329de

	def _plot_equity_section():
	"""Equity section"""
	# Max DD Dur. line
	equity = equity_data['Equity'].copy()
	dd_end = equity_data['DrawdownDuration'].idxmax()
	if np.isnan(dd_end):
	dd_start = dd_end = equity.index[0]
	else:
	dd_start = equity[:dd_end].idxmax()
	# If DD not extending into the future, get exact point of intersection with equity
	if dd_end != equity.index[-1]:
	dd_end = np.interp(equity[dd_start],
	(equity[dd_end - 1], equity[dd_end]),
	(dd_end - 1, dd_end))

backtesting.py/backtesting/_plotting.py

Lines 354 to 359 in 05329de

	dd_timedelta_label = df['datetime'].iloc[int(round(dd_end))] - df['datetime'].iloc[dd_start]
	fig.line([dd_start, dd_end], equity.iloc[dd_start],
	line_color='red', line_width=2,
	legend_label='Max Dd Dur. ({})'.format(dd_timedelta_label)
	.replace(' 00:00:00', '')
	.replace('(0 days ', '('))

I don't suppose something strikes you as odd at first glance? 😅

[eervin123]

backtesting.py/backtesting/lib.py

Lines 62 to 66 in 05329de

 _EQUITY_AGG = { 
     'Equity': 'mean', 
     'DrawdownPct': 'max', 
     'DrawdownDuration': 'max', 
 } 

('Equity' should maybe aggregate with 'last'. 🤔)

That did the trick. Here is the before picture

with a zoom in on the problem period.

As you can see above, this is very close and likely the culprit from some downsampling throwing an erroneous peak.

Here is an after picture after I set the 'Equity' : 'last' in the _EQUITY_AGG dict.

The Max Drawdown Duration perfectly matches what is expected from the _compute_stats calculations.

[kernc]

Pleasure to work with you.