Skip to content

Glossary

Savvas Leousis edited this page Sep 26, 2026 · 2 revisions

Glossary

Plain meanings of the metrics the console shows and the words used in Stonks. Every "?" tip in the console links here.

flowchart LR
  BAR[bar] --> LAKE[lake] --> LAB[lab] --> SH[shadow] --> AC[active] --> TICK[tick]
Loading

Returns and risk

Metric Meaning
Total return How much the value grew or shrank over the whole period, as a percentage.
CAGR Compound annual growth rate: the steady yearly growth that gives the same end value.
Equity Total value of the account: cash plus everything it holds at current prices.
Volatility How much returns swing around their average, scaled to a year. Higher means bumpier.
Sharpe ratio Average return divided by volatility, scaled to a year. Return per unit of bumpiness.
Sortino ratio Like Sharpe, but only counts downside swings, so big up days are not penalised.
Calmar ratio CAGR divided by the worst drawdown: growth earned per unit of pain.
Drawdown How far value has fallen below its previous peak.
Max drawdown The deepest fall from a peak during the period.
Drawdown duration How many bars the longest stretch below a previous peak lasted.
Ulcer index Measures both how deep and how long drawdowns are; lower is calmer.
UPI (Ulcer performance index) CAGR divided by the Ulcer index.
VaR (value at risk) A bad-day loss: on 95% of days the loss is smaller than this.
ES (expected shortfall) The average loss on the days worse than VaR. Also called CVaR.
Skew Whether big moves tend to be up (positive) or down (negative).
Kurtosis How often extreme moves happen compared with a normal curve (normal is 3).
Exposure Share of the account invested rather than held as cash.
Turnover How much of the portfolio is bought and sold per year. More turnover means more costs.
Slippage Paying a worse price than expected when an order fills.
Basis point (bps) One hundredth of a percent: 25 bps is 0.25%.

Trades

Metric Meaning
Trades Completed round trips (buy then sell). Few trades make every statistic shaky.
Win rate Share of trades that made money. A low win rate can still pay if winners are big.
Profit factor Money won on winners divided by money lost on losers. Above 1 is profitable.
Expectancy The average profit or loss per trade, winners and losers together.
Payoff ratio Average winner divided by average loser.
Risk of ruin Chance that a bad run of trades causes a drawdown beyond a set limit (40% by default).

Against a benchmark

Metric Meaning
Benchmark What the strategy is compared with: SPY, an equal-weight universe or a ticker you pick.
Beta How much the strategy moves with the benchmark. 1 means in step, 0 means unrelated.
Alpha Return the benchmark does not explain: what is left after accounting for beta.
Tracking error How far returns stray from the benchmark, scaled to a year.
Information ratio Extra return over the benchmark divided by tracking error. Skill per unit of difference.

Validation and statistics

Metric Meaning
In-sample (IS) score The objective on the data the tuner used. Usually too optimistic.
Out-of-sample (OOS) score The objective on data the tuner never saw: the honest number.
Best score The best objective value the tuner found (for example Sharpe), before survival tests.
Probabilistic Sharpe (PSR) The chance the true Sharpe is above zero, given how long and how bumpy the record is.
Deflated Sharpe (DSR) PSR that also accounts for how many variants were tried. Many tries make a lucky best more likely.
MinTRL (minimum track record length) How long a record must run before its Sharpe can be trusted.
PBO (probability of backtest overfitting) How often the best variant in one half of the data is below average in the other half. Lower is better.
p-value The chance of a result this good from luck alone. Small (under 0.05) means probably not luck.
MCPT (Monte Carlo permutation test) Shuffle the bars many times; the real result should beat most shuffles.
Walk-forward Tune on the past, test on the next slice, roll forward and repeat.
Walk-forward efficiency (WFE) Out-of-sample return divided by in-sample return across walk-forward folds. Near 1 means little overfitting.
Embargo Bars skipped between the train and validation windows so information cannot leak across.
Period stability Checks the strategy works in each sub-period, not just in one lucky stretch.
Perturbation Adds small noise to prices to see if the results hold.
Plateau Nudging each parameter a little should cost little; a sharp peak is a warning.
Cost stress Rerun with costs doubled or more; the break-even multiple is where the edge disappears.
Drift A change in how the data behaves between the training period and later ones.
PSI (population stability index) How much a distribution shifted. Under 0.1 is stable; over 0.25 is a big shift.
Runs test Checks whether wins and losses cluster together more than chance would.
IC (information coefficient) Rank correlation between a signal today and the returns that follow. Small positive values are normal.
ICIR Average IC divided by its standard deviation: how steady the signal's edge is.
Event study Compares returns after the strategy's entries with just holding the same tickers.
Survival test A check that tries to break a strategy before it is trusted.

Stonks words

Term Meaning
Active Registry status of a strategy the tick trades.
Asset class equity, crypto, commodity or bond.
Backtest Replaying history to see how a strategy would have done.
Bar One period of prices: open, high, low, close, volume.
Book One portfolio's set of strategies, constructor, risk policy and broker in a tick.
Client id Deterministic order id built from the date, portfolio, strategy, ticker and side.
Constructor Turns strategy signals into target positions (single winner, equal weight, vol target, ...).
Dry run A tick that scores and logs but places no orders.
Go-live check The checks a paper record must pass before promotion.
Idempotent Safe to repeat: running twice gives the same result as once.
Ingest Pulling data from a vendor into the lake.
Interval Bar size: 1m, 5m, 15m, 30m, 1h, 4h, 12h, 1d, 1w.
Lake The DuckDB file with market data (data/lake.duckdb).
Long-only Can buy and sell what it holds, never short.
Look-ahead Using data that was not available yet at that time. A bug that fakes good results.
MCP Model Context Protocol; lets AI assistants call tools.
Mode How a subscription acts: notify, paper or auto.
Model book A shadow strategy's virtual portfolio.
Paper trading Trading with fake money, in Stonks or at a real broker's paper account.
Point-in-time Using only data that was published by that date.
Preset A named set of survival tests: quick, standard, promotion.
Quarantine Keeping impossible bar rows out of the lake, in quarantined_bars.
Reconcile Syncing order status and fills from the broker into state.
Registry Where strategies and their status are stored.
Risk rule A rule that clips or drops orders before the broker.
Seam An interface where new behavior plugs in.
Shadow Registry status: evaluated each tick on a virtual portfolio, never traded.
Soft-fail Log the error, skip that item, keep going.
State The SQLite file with trading records (data/state.sqlite).
Subscription A user following a strategy in a mode, for a portfolio.
Tick One run of the trading loop, usually daily.
Tuner Searches parameter values for the best score.
Universe The list of tickers a strategy or tick looks at.
Upsert Insert a row, or update it if it already exists.

Clone this wiki locally