Account report total cash divergence with manually calculated equity

[Egisess]

While trying to calculate final equity encountered strange behavior:

Engine Logs

account = engine.trader.generate_account_report(BINANCE)

tmp = pd.DataFrame(account['total'])
tmp['total'] = (tmp['total'].astype('float') - 1e6) # Example strategy works with initial 1e6 USD
tmp.plot()

We see here reduce of total cash, but the engine logs states that we have positive PnL:

2025-05-21T10:55:52.513326003Z [INFO] BACKTEST_TRADER-001.BacktestEngine: -----------------------------------------------------------------
2025-05-21T10:55:52.513326004Z [INFO] BACKTEST_TRADER-001.BacktestEngine:  PnL Statistics (USD)
2025-05-21T10:55:52.513327000Z [INFO] BACKTEST_TRADER-001.BacktestEngine: -----------------------------------------------------------------
2025-05-21T10:55:52.513412000Z [INFO] BACKTEST_TRADER-001.BacktestEngine: PnL (total):                    0.037500000000000006
2025-05-21T10:55:52.513412001Z [INFO] BACKTEST_TRADER-001.BacktestEngine: PnL% (total):                   3.7499999976716937e-06
2025-05-21T10:55:52.513413000Z [INFO] BACKTEST_TRADER-001.BacktestEngine: Max Winner:                     1.83
2025-05-21T10:55:52.513413001Z [INFO] BACKTEST_TRADER-001.BacktestEngine: Avg Winner:                     0.4491304345652174
2025-05-21T10:55:52.513413002Z [INFO] BACKTEST_TRADER-001.BacktestEngine: Min Winner:                     0.03
2025-05-21T10:55:52.513413003Z [INFO] BACKTEST_TRADER-001.BacktestEngine: Min Loser:                      0.0
2025-05-21T10:55:52.513413004Z [INFO] BACKTEST_TRADER-001.BacktestEngine: Avg Loser:                      -0.31973684184210527
2025-05-21T10:55:52.513413005Z [INFO] BACKTEST_TRADER-001.BacktestEngine: Max Loser:                      -1.13999999
2025-05-21T10:55:52.513413006Z [INFO] BACKTEST_TRADER-001.BacktestEngine: Expectancy:                     0.10130952380952388
2025-05-21T10:55:52.513414000Z [INFO] BACKTEST_TRADER-001.BacktestEngine: Win Rate:                       0.5476190476190477
2025-05-21T10:55:52.513414001Z [INFO] BACKTEST_TRADER-001.BacktestEngine: -----------------------------------------------------------------
2025-05-21T10:55:52.513414002Z [INFO] BACKTEST_TRADER-001.BacktestEngine:  Returns Statistics
2025-05-21T10:55:52.513414003Z [INFO] BACKTEST_TRADER-001.BacktestEngine: -----------------------------------------------------------------
2025-05-21T10:55:52.516823000Z [INFO] BACKTEST_TRADER-001.BacktestEngine: Returns Volatility (252 days):  0.005356806978984849
2025-05-21T10:55:52.516824000Z [INFO] BACKTEST_TRADER-001.BacktestEngine: Average (Return):               0.0005762588436168704
2025-05-21T10:55:52.516824001Z [INFO] BACKTEST_TRADER-001.BacktestEngine: Average Loss (Return):          -0.0010018876746293681
2025-05-21T10:55:52.516824002Z [INFO] BACKTEST_TRADER-001.BacktestEngine: Average Win (Return):           0.0017850519214225
2025-05-21T10:55:52.516824003Z [INFO] BACKTEST_TRADER-001.BacktestEngine: Sharpe Ratio (252 days):        0.11199800894198363
2025-05-21T10:55:52.516824004Z [INFO] BACKTEST_TRADER-001.BacktestEngine: Sortino Ratio (252 days):       nan
2025-05-21T10:55:52.516825000Z [INFO] BACKTEST_TRADER-001.BacktestEngine: Profit Factor:                  2.3260935451974576
2025-05-21T10:55:52.516825001Z [INFO] BACKTEST_TRADER-001.BacktestEngine: Risk Return Ratio:              0.31166158558397655

Engine get_result()

Furthermore, If we try to regenerate report:

tmp = engine.get_result()
tmp.stats_pnls

We get

{'USD': {'PnL (total)': -0.1625,
  'PnL% (total)': -1.624999999767169e-05,
  'Max Winner': 1.83,
  'Avg Winner': np.float64(0.4491304345652174),
  'Min Winner': np.float64(0.03),
  'Min Loser': np.float64(0.0),
  'Avg Loser': np.float64(-0.31973684184210527),
  'Max Loser': np.float64(-1.13999999),
  'Expectancy': np.float64(0.10130952380952388),
  'Win Rate': 0.5476190476190477}}

That looks nothing like we've seen above in strategy logs.

Manual PnL calculation

report = engine.trader.generate_positions_report()
report = report.set_index('ts_closed')
sns.lineplot(report["realized_pnl"].apply(lambda s: float(s.split()[0])).cumsum())
plt.xticks(rotation=45)
plt.show()

And this result seems the most reliable.

So the question is: Why all these seemingly similar operations give such drastically divergent results?
In my opinion last iteration of PnL calculation is the most accurate but the question is: Why the equity growth is not reflected in account report total balance?

[cjdsellers]

Hi @Egisess

Thanks for reaching out with the detailed report.

I managed to track it down to position flip scenarios for margin accounts.

Now fixed from d3abd99 which adds tests to confirm position PnL exactly matches account PnL, including your example inputs.

Please confirm the fix on your end what you have the chance.

[Egisess]

Thank you for paying such close attention to my report