14.10 Going live · M14 Portfolio and going live

14.10.1How many trades distinguish edge from noise#

The derivation, the false-positive rates and the selection trap are in M03 · 3.1.1, where they belong mathematically (next to expectancy and the SQN). The two results are restated here with M14 numbers because the graduation rules below use them.

Proposition 14.16= Corollary 3.1.1

With i.i.d. R-multiples of mean $E > 0$ and standard deviation $s$ , the expected $t$ -statistic $E\sqrt N/s$ reaches 2 when

N \ge \left(\frac{2s}{E}\right)^2. \tag{14.8}

Proposition 14.17= Corollary 3.1.2

Under the normal approximation, $P(t \ge 2) \ge 1 - \beta$ requires

N \ge \left(\frac{(2 + z_{1-\beta})\,s}{E}\right)^2, \qquad z_{0.8} = 0.8416. \tag{14.9}

For a typical swing setup ( $E = 0.20R$ , $s = 1.50R$ ) that is 225 trades for $t = 2$ in expectation and 455 for 80% power (M03 · 3.1.1 tabulates three setups). At three trades a week, 225 trades take 75 weeks, so paper trading cannot prove a swing edge. The design that works: establish the edge on historical data with the M04 safeguards (out-of-sample, walk-forward, deflated Sharpe), then use paper and small live size to check that live results are consistent with the backtest distribution (14.10.3), which needs far fewer trades.

14.10.2A template trading plan#

Write it before trading, sign and date it, and change it only at the monthly review.

Purpose and capital. Capital allocated to active trading (a fixed sum you can lose entirely without changing your life or your company's runway). Long-term savings stay in index funds (M06 · 6.11).
Markets and instruments. E.g. NSE cash equities in the Nifty 200; no options buying until the options section of the plan is written (M09).
Setups. A finite list. For each: entry rule, stop rule, exit rule, holding period, the backtest statistics ( $E$ , $s$ , $N$ , deflated Sharpe from M04), and the market regime in which it is allowed.
Sizing (from M03). Risk $f$ per trade as a fraction of current equity: start at 0.25–0.5%, with a ceiling well below fractional Kelly given estimation error. Shares $= fE/(P_{entry} - P_{stop})$ , rounded down. Notional cap per name.
Portfolio limits. Maximum open risk $H$ ; maximum positions per sector; maximum gross exposure; stress-test rule.
Loss limits. Daily loss limit $L_d$ and weekly $L_w$ in R and in rupees, calibrated from the backtest's daily-loss distribution (14.8). When hit: flat, no new trades until the next period, journal entry required.
Drawdown rule. At drawdown $d_1$ halve $f$ ; at $d_2$ stop and return to paper; restart conditions written in advance.
Execution and costs. Order types (M01), cost model per trade, slippage assumption; compare live slippage monthly.
Routine. Pre-market checklist; no new positions in the first 15 minutes unless the setup specifies it; event calendar check.
Journal and review cadence. Fields of 14.9.5; weekly and monthly reviews; quarterly plan review.
Graduation and scaling criteria (next section).
Compliance and tax. Retail algo rules if automating (the SEBI retail algorithmic-trading framework and the exchanges' implementation circulars; dated status in M01's Facts table); tax treatment of intraday (speculative business income) versus F&O (non-speculative business income) versus delivery (capital gains), as of FY 2026-27 (M00 · 0.7); check with a chartered accountant.
Operational controls if any order is sent by code. Write these down as rules, and test each one on paper before going live:
- Kill switch: one action that cancels all open orders and stops the program, reachable without the program itself (the broker's terminal or phone line), and tested monthly.
- Order-rate and size caps in the program, below the broker's and the exchange's limits: maximum orders per second and per day, maximum quantity and notional per order (at most the plan's per-name cap), and a hard stop when the daily loss limit (item 6) is hit.
- Idempotent orders: every order carries a client order id derived from (strategy, signal date, instrument, leg), so that a retry after a time-out cannot create a second position; the program checks the order book for that id before re-sending.
- Broker or API outage: a written fallback (who places the exit orders manually, from which terminal, at what price limits), and the rule that the program does not open new positions while it cannot see its open positions.
- End-of-day reconciliation: program orders, fills and positions against the broker's contract note and against the exchange's trade confirmation sent by SMS/e-mail to the registered contact; any mismatch stops the next session's trading until explained.

14.10.3Paper-trading protocol and graduation#

Protocol

Trade only the plan's setups, in the plan's size (in notional rupees, as if real), recording fills at prices you could actually have obtained (use the next tick or the ask for buys, not the last trade).
Charge full costs (M01) on every trade.
Record every signal, including those you did not take; skipped signals are a rule-adherence failure.

Graduation from paper to small live size requires all of:

At least $N_0$ trades (e.g. 50–100), enough to estimate rule adherence and slippage.
Rule adherence ≥ 95% of trades.
Live mean R within the backtest's sampling range: $|\bar R_{live} - E_{bt}| \le 2s/\sqrt{N_0}$ . This is a consistency test, not a proof of edge; the proof was the backtest.
Realised costs and slippage within the cost model's assumptions.
No loss-limit or drawdown-rule breach in the last month.

Graduation from small to full size: the same criteria on live trades, plus a cumulative live $t$ -statistic consistent with the backtest. Scale in steps (for example 25% → 50% → 100% of the plan's $f$ ), each step after another $N_1$ live trades meeting the criteria. Scale down immediately on a drawdown-rule trigger. Never scale up after a winning streak alone: by Douglas's third truth, streaks are what random sequences look like.