Risk and sizing¶

The idea¶

You have been choosing what to trade. This level is about how much, and it matters more.

Two traders run the identical rule on the identical instrument over the identical year. One finishes up 15%, the other is wiped out. The only difference is position size. There is no entry signal good enough to survive bad sizing, and there are mediocre signals that make money under good sizing.

The question to answer before every trade is not "how much can I make?" but "what does this cost me if I am wrong?"

Why it matters¶

Drawdown arithmetic is not symmetric. A loss needs a larger gain to undo it, and the gap widens fast:

Fall from peak	Gain needed to recover
10%	11%
20%	25%
33%	50%
50%	100%
75%	300%
90%	900%

At 50% down you need to double to get back to where you started. This is why Arvo's default verdict criteria include a drawdown under 30% — past that, the recovery required stops being a trading problem and becomes a fantasy.

Losing streaks are normal, not evidence of failure. A rule that wins 40% of the time will hit six losses in a row about once every two hundred trades. Not if something goes wrong — as ordinary arithmetic. Your sizing has to be survivable through a streak you should expect, and most beginners size for the average case and are destroyed by the ordinary one.

What it looks like¶

The standard formula, and it is the whole of position sizing:

position size = (capital × risk per trade) ÷ (stop distance per unit)

Worked, with $50,000, risking 1% per trade, on a stock at $100 whose ATR-based stop sits $4 below:

risk budget     = $50,000 × 0.01  = $500
stop distance   = $4 per share
position size   = $500 ÷ $4       = 125 shares  ($12,500 of stock)

Note what happened: a wider stop buys fewer shares. The risk stays $500 either way. This is the mechanism that makes a $600 stock and a $20 stock comparable, and it is why stop distance is measured in ATRs rather than dollars or percent.

Note also what is not in the formula: your confidence, how good the setup looks, or how the last trade went.

What people get wrong¶

Sizing by capital instead of by risk. "I put 10% of my account in each position" sounds disciplined and is not — 10% of your account in a quiet utility and 10% in a volatile small cap are completely different risks.

Adding to a loser. Averaging down increases position size exactly as the evidence against the position accumulates.

Ignoring correlation. Five positions in five different semiconductor companies is one position in semiconductors, sized five times. The diversification is nominal, and the loss arrives all at once.

Trading bigger after a loss to get it back. Arithmetic does not care about your intentions, and this is how a survivable drawdown becomes a terminal one.

See it in Arvo¶

Arvo does not let sizing be an afterthought: every entry is a proposal, and it passes one gate that sizes it, reduces it, or refuses it with a named reason. The same gate runs in the backtest and live, so a stored verdict describes a system that actually exists. Set in risk.json, or Arvo's defaults when there is none:

Field	What it limits
`stop_atr_multiple`	stop distance as a multiple of ATR. Not set runs with no stop
`risk_per_trade`	fraction of starting capital a stop-out may cost. Needs a stop
`max_position_fraction`	fraction of capital one position may occupy. Above 1 is leverage
`max_drawdown`	fall from peak that stops the rule for the run. Permanent
`max_concurrent_positions`	the most positions held at once
`max_daily_loss`	realised loss in one day that refuses new entries
`correlation_cap`	above this pairwise correlation, names are one bet. Refuses when unknown
`sector_cap`	the most positions one sector may hold. Refuses an unlabelled name
`day_trading`	`pattern_day_trader`: three round trips per five days under $25k

Three behaviours worth knowing, because they are unusual:

A refusal is a fact, not a log line. NoStop, TooSmall, NotWholeLot, DailyLossLimit, TooManyPositions, Correlated, Halted — each named on the record and counted. "Why didn't it take that trade?" is answerable.
The warning band at four fifths. At 80% of any limit the gate says so before it acts, writes a warning event and raises an alert. This is the one moment a person can act before the gate acts for them.
The drawdown halt is permanent for the run and lifts for nobody. The kill switch is manual and lifts only when a person releases it. A session that starts against an account already holding positions adopts them and comes up halted until you look — a limit that a restart lifts is not a limit.

Full field reference: the risk model. Why it is one gate rather than two: the risk gate.

Watch¶

(to be chosen) — position sizing worked through from account to share count.
(to be chosen) — the drawdown table above, and why recovery is harder than the fall.

Next¶

Strategy families — the five shapes Arvo can test, one lesson each.