Reconstruct realized P&L

Setup. The scripts sit beside this file. Run them from a working folder that holds your .env (E*TRADE consumer key/secret) and a data/ directory; install deps with pip install -r requirements.txt where one is included. Part of a set: etrade-pull, order-check, portfolio-review, realized-pnl, rebalance-plan, price-history, trade-journal, intrinsic-value, sec-financials, buffett-checklist.

Needs data/transactions.csv first — see etrade-pull.

python realized_pnl.py        # closed lots, best/worst, by-symbol roll-up
python backtest_holding.py    # for every loss taken: what if it had been held?

load.py does the normalizing (epoch → datetime, trades split from cash) and is importable: from load import trades, cash.

Network access

realized_pnl.py and load.py are offline: they read only data/transactions.csv. Only backtest_holding.py goes online, through etrade.py, to fetch live quotes. It talks to E*TRADE's own API and nothing else:

host	used for
api.etrade.com	market quotes (/v1/market/quote), OAuth request/access tokens
apisb.etrade.com	the same, against E*TRADE's sandbox when ETRADE_ENV is not prod
us.etrade.com	the authorize page you open in your browser to approve access

It reads only your E*TRADE consumer key/secret and tokens from your .env, signs requests with them, and sends nothing else anywhere: no telemetry, no third-party host.

Rules

• The in-window total is not lifetime performance. Sells whose matching buys predate the ~3-year API window come back in a separate orphan_sells table with no cost basis. They are usually the oldest and largest winners, so quoting the net realized figure as "how they've done" can invert the sign of the answer. Report the closed-lot total and the orphan count, always together.

• FIFO is an assumption, not their tax treatment. The matcher pairs each sell against the oldest open buy. If the broker used specific-lot identification, the per-lot gains here won't match the 1099. Fine for reading behavior, wrong for filing. Say which one you're doing.

• Dividends carry a symbol but are not trades. load.py keys off transactionType, not the presence of a ticker, for exactly this reason. If you filter the raw CSV yourself, don't reintroduce the bug.

The counterfactual is usually the finding

backtest_holding.py answers one question: for every lot sold at a loss, what would it be worth held to today, and at its 52-week high?

This is the highest-value output in the repo for most people, because a book can look healthy on holdings and still be bleeding from a repeated exit mistake. That pattern is invisible in a positions table.

Read it carefully:

• It ignores redeployment, dividends and taxes. It isolates one question — did cutting on price cost or save money — and answers only that. The capital didn't sit in cash in real life, so this is not a portfolio-level P&L claim.
• It needs live quotes, so it hits the API. Auth applies: etrade-pull.
• A large, consistent counterfactual against the person is a behavioral finding. Surface it to portfolio-review as such, and pair it with the temperament section of buffett-checklist rather than a list of the individual trades.
• One or two bad exits is not a pattern. Look for direction across most of the losing names.