All projects
BuildingCarver (systematic risk) · Damodaran (valuation) · Kelly (1956)

Practice instruments — ten simulators for the arithmetic of risk

Most explanations of position sizing are a paragraph and a formula. The formula is memorable and the consequence is not — it only lands when you can hold the edge fixed, move the bet size, and watch the same two thousand worlds end differently. Ten linked instruments: Kelly and sizing, small-sample horizons, compounding drag, volatility targeting, portfolio risk decomposition, calibration, valuation, options, thesis falsifiability, and a Brier-scored drill that tests whether any of it stuck.

Related research

Knowledge Base
AtlasMarkets

Where it runs

In-labRuns entirely in your browser — no server, no accounts, nothing stored. Every simulation is seeded, so the same inputs always give the same worlds.

Live demo

Kelly & position sizing

A real edge, sized wrong, still ends at zero.

Presets

Each one makes a specific argument.

The process

What is actually true about your edge.

60.0%

A 52% win rate at even money is a realistic ceiling for a skilled discretionary trader.

: 1

Your estimate

The edge you think you have — where the real damage starts.

Off — you are assumed to know your edge exactly. Nobody does.

Sizing

The only variable most traders set last.

1.00x

Betting 0.00% of wealth per bet.

The run

Same seed, same worlds.

Dead below 1.0% of stake.

Expected log growth
never
Median outcome
what a typical path got
Mean outcome
flattered by a few survivors
Ruined
no path hit the floor

Expected log growth against bet size

The peak is Kelly. The second zero is the cliff. Everything red is an edge you really have, sized into certain ruin.

f* 0.00%
cliff

The same worlds, this bet size

2,000 paths over 250 bets. Change the sizing and these are the identical 2,000 worlds — only your decision moves.

median path
mean wealth
ruin floor

Where it all ended

Log-binned, because a linear histogram of wealth is one bar at zero.

What actually happened

The numbers behind the picture.

Ended below start
Lost over half
Median worst drawdown
the typical path's worst moment
95th pct drawdown
the one that ends careers
5th pct outcome
95th pct outcome

Reality check — the edge you think you have

Overestimating your edge does not scale your bet a little. It scales it by the ratio of the two Kelly fractions.

Kelly you believe
0.00%
off your estimated edge
Kelly in reality
no edge
off the true edge
You actually bet
0.00%
1.00x of believed Kelly
True Kelly multiple
what you are really running
Estimation error is off — you are assumed to know your edge exactly. Turn it on to see the gap between the edge you think you have and the one you do.

What if I'd sized differently

One Monte Carlo per multiple, across the whole range.

Runs 40 separate simulations. Takes a second or two.

Simulated results from a seeded pseudo-random generator. Nothing here is a forecast, a recommendation, or a claim about any real instrument — it is a machine for developing intuition about the arithmetic of risk. Anything you type stays in your browser.

Stack

TypeScriptReactCanvasMonte Carlo

The problem

Sizing is the decision most traders make third, after entry and exit. It is the one that determines whether the other two ever get to matter — and the arithmetic that shows why is genuinely hard to believe until you watch it happen.

The approach

Pure-TypeScript simulation core with no framework dependencies, hand-drawn canvas charts, and 180 tests pinning every numeric claim the interface makes. Where a closed form exists it is used instead of a simulation: the terminal distribution after n bets is exactly binomial, so nothing about it needs sampling.

Decisions & trade-offs

  1. 01Every path gets its own seeded RNG, so moving the sizing slider re-runs the SAME worlds with one decision changed. A shared generator would confound every comparison with luck.
  2. 02Yes/no drill questions are graded on the Brier score, not on correctness — marking on correctness would train exactly the overconfidence the Kelly instrument shows turning into 5x sizing.
  3. 03Drill questions are generated FROM the simulators, so answers are computed rather than authored and the bank cannot drift out of sync with the maths.
  4. 04No worked company case ships with it. Constructing one from the filings is where the learning is; the valuation instrument reads a CSV you author instead.

Where it stands

Live and compounding. The instruments already produced findings worth keeping: believing you win 60% while actually winning 52% puts full Kelly at exactly 5x true Kelly; a genuine 52/48 edge sized correctly finishes underwater 57% of the time over ten bets; and six positions at 0.6 average correlation behave like one and a half.

Build log

Running diaryNotebook

The running diary appears here as /notebook entries once the build starts.

Practice instruments — ten simulators for the arithmetic of risk · Lab | MenFem