Skip to content

Real-Trade Review

A backtest report shows theoretical positions; a real-trade review shows what you actually traded. TradeReview uses actual fill prices, fees, taxes, dividends and deposits/withdrawals to rebuild daily holdings, cash and NAV, computes time-weighted returns that exclude external cash flows, and produces the same FinLab report you get from a backtest.

Review straight from a broker account

from finlab.online.schwab_account import SchwabAccount
from finlab.portfolio import TradeReview

acc = SchwabAccount()
review = TradeReview.from_account(acc, start="2026-01-01")

review.summary()  # start/end NAV, net inflow, dividends, fees and tax, P&L, TWR
review.daily  # daily cash, market value, NAV, flows, return, cumulative P&L
review.holdings  # shares held at each close
review.trades  # FIFO-matched round trips (with dividends); open lots have no exit_date
review.limitations  # data limitations of this review: always read them
review.report.display()  # FinLab report: equity curve, metrics, MAE/MFE, trades

from_account:

  1. Pulls fills from start to today through the broker API (see the table below).
  2. Rebuilds starting holdings from the broker's current holdings (undoing fills, stock dividends and splits in the window), then reconciles the rebuilt ending holdings with the broker; mismatches are listed in review.limitations.
  3. Applies dividends: Schwab and Binance report them in the account records; for Taiwan accounts FinLab applies its dividend announcements automatically (cash dividends on the ex-dividend date, stock dividends and ETF splits on the ex-rights / resumption date).
  4. Values holdings daily at unadjusted close.

Two return modes

Mode When NAV External flows
Account Broker API has complete deposits/withdrawals (Schwab, Binance), or you pass initial_cash and cash_events cash + market value deposits, withdrawals
Holdings Taiwan brokers (the settlement account is a bank account the API cannot see) market value buys are invested before the open; sell proceeds and dividends are withdrawn after the close

Daily return r = (NAV + outflow) / (previous NAV + inflow) - 1, chained into the equity curve; CAGR, Sharpe and drawdown in the report are computed from it and are unaffected by deposits and withdrawals. In account mode, pnl is the NAV change minus net external inflow.

Brokers without deposit/withdrawal data (the Taiwan brokers) review holdings only. To get a full account NAV for such an account, pass starting cash and cash flows:

import pandas as pd

review = TradeReview.from_account(
    acc,
    start="2026-01-01",
    initial_cash=50_000,
    cash_events=pd.DataFrame(
        {
            "time": ["2026-03-02"],
            "amount": [10_000],
            "type": ["deposit"],
        }
    ),
)

Broker support and limitations

Broker Fill source Fees / tax Closed positions Dividends Deposits / withdrawals Default mode
Schwab (U.S.) get_transactions net amount minus gross yes broker records broker records account
Sinopac (Taiwan) realized P&L + detail, open lot detail buys from detail; sells derived from realized P&L (combined in fee) buy legs recovered FinLab auto none holdings
Pocket (Taiwan) realized P&L, open lot detail none as Sinopac when the SDK has the detail call; otherwise shares sold in the window are treated as held at start FinLab auto none holdings
Fubon (Taiwan) filled_history (30 days per call, fubon_neo 2.1.1+) not in API; estimated from fee_ratio and statutory tax yes FinLab auto none holdings
Masterlink (Taiwan) statement account_statment actual per fill yes FinLab auto none holdings
E.SUN (Taiwan) get_transactions_by_date (180 days per call) actual per fill yes FinLab auto none holdings
Binance myTrades (per symbol) paid in quote or base asset; BNB-paid fees are not counted only currently held assets and those passed in symbols= quote-asset distributions quote-asset deposits/withdrawals account

Common limitations:

  • Only long cash and margin positions are reviewed; short-sale and securities-lending fills are skipped and listed in limitations. Margin interest is not counted.
  • Fills that only have a date (most Taiwan brokers) are processed buys first, then sells, on the same day, which does not affect end-of-day valuation.
  • Corporate actions the broker or FinLab does not supply are not applied automatically; add them to corporate_actions (columns stock_id, date, cash, ratio). Taiwan capital reductions and par value changes fall in this group; Taiwan stock dividends are converted with the par value, rounding fractional shares down.
  • When a broker API cannot return data that far back, the review can only start from the earliest available date.

Use your own fills

For other brokers or exported files, map them to the standard columns and build the review directly:

import pandas as pd
from finlab.markets.us import USMarket
from finlab.portfolio import TradeReview

fills = pd.DataFrame(
    {
        "time": ["2026-01-05 09:45", "2026-01-07 10:00"],
        "stock_id": ["MSFT", "MSFT"],
        "action": ["buy", "sell"],
        "quantity": [50, 50],  # shares
        "price": [480.0, 486.5],
        "fee": [0, 0],  # actual amount of this fill; missing fee/tax columns count as 0
    }
)
market = USMarket()
review = TradeReview(
    fills,
    market.get_price("close", adj=False),
    start="2026-01-05",
    initial_holdings={"AAPL": 100},
    # Illustrative dividend: cash per share on the ex-date; ratio 2.0 would be a 2-for-1 split
    corporate_actions=pd.DataFrame(
        {"stock_id": ["AAPL"], "date": ["2026-02-09"], "cash": [0.25], "ratio": [1.0]}
    ),
    market=market,
)

For Taiwan fills, finlab.portfolio.trade_review.tw_corporate_actions(stock_ids, start, end) builds the corporate_actions table from FinLab's Taiwan dividend announcements.

Compare with strategy targets

compare uses the actual position at the close of each target date, never later fills:

target = pd.DataFrame({"AAPL": [20], "MSFT": [10]}, index=["2026-03-02"])
review.compare(target)  # date, stock_id, target, actual, diff (shares)

weights = pd.DataFrame({"AAPL": [0.6], "MSFT": [0.4]}, index=["2026-03-02"])
review.compare(weights, kind="weight")  # or compare weights