Python bet sizing and bankroll simulation for developers modeling repeated binary outcomes.
Given your estimated win probability, payoff, loss, and transaction cost, Keeks
returns the fraction of your bankroll to stake — and can simulate that same rule
over repeated trials to show the bankroll path it produces. For portfolios of
options described by return distributions rather than single bets, the
keeks.allocation layer
sizes them together and returns one long-only weight per option.
pip install keeksKeeks supports Python 3.10 through 3.14. The portfolio-allocation layer's
QP/LP solvers need one optional extra (everything else is numpy-only):
pip install "keeks[allocation]".
from keeks.binary_strategies import KellyCriterion
bankroll = 1_000.0
strategy = KellyCriterion(
payoff=1.0,
loss=1.0,
transaction_cost_rate=0.01,
)
fraction = strategy.evaluate(probability=0.55, current_bankroll=bankroll)
amount = bankroll * fraction
print(f"Bankroll fraction: {fraction:.4%}")
print(f"Amount from a $1,000 bankroll: ${amount:.2f}")Bankroll fraction: 9.0009%
Amount from a $1,000 bankroll: $90.01
evaluate() returns a fraction, not a currency amount; the example multiplies it by the bankroll only to make the result concrete.
Binary strategies size one bet at a time; the allocation layer sizes a whole
portfolio at once, returning one long-only weight per option (each in [0, 1],
summing to at most one — the residual is cash held at zero return). Adapters
build a joint-return model from what you have — binary bets, a scenario matrix,
fitted marginals, or any callable sampler — and allocators consume it through
one sampling contract. MeanVariance at risk aversion λ = 1 is the
second-order Kelly allocation; the
allocation documentation
covers the other method families (scenario-based, hierarchical, online, and the
covariance estimators).
from keeks import MeanVariance, binary_bets_model
# Three simultaneous binary bets: (win probability, decimal odds, loss)
model = binary_bets_model([(0.55, 2.0, 1.0), (0.42, 2.5, 1.0), (0.70, 1.5, 1.0)])
# λ = 1 mean-variance is the second-order Kelly allocation
strategy = MeanVariance.from_model(model, n_samples=4096, seed=42)
print(strategy.evaluate(current_bankroll=1_000.0)) # weights, not amountsAll nine strategies expose evaluate(probability, current_bankroll), but their
constructors and sizing rules differ.
| Strategy | Choose it when you want to model |
|---|---|
KellyCriterion |
Full Kelly sizing from a binary win probability, payoff, loss, and cost. |
FractionalKellyCriterion |
A fixed fraction of the full-Kelly result. |
DrawdownAdjustedKelly |
Kelly sizing scaled by an acceptable-drawdown input. |
OptimalF |
A geometric-growth rule based on a supplied win rate, with a risk-fraction cap. |
FixedFractionStrategy |
A constant fraction above a minimum probability; useful as a baseline. |
CPPIStrategy |
A cushion-based rule relative to a bankroll floor. |
DynamicBankrollManagement |
A fraction adjusted from recent recorded outcomes. |
MertonShare |
A CRRA risk-aversion rule adapted to binary outcomes. |
NaiveStrategy |
A positive-expected-value rule without utility-based sizing. |
See the strategy API for constructor parameters and formulas.
Once you have chosen a strategy, pass it and a fresh BankRoll to a simulator:
from keeks.bankroll import BankRoll
from keeks.binary_strategies import FractionalKellyCriterion
from keeks.simulators.repeated_binary import RepeatedBinarySimulator
bankroll = BankRoll(
initial_funds=1_000.0,
percent_bettable=0.8,
max_transaction_loss=0.3,
)
strategy = FractionalKellyCriterion(
payoff=1.0,
loss=1.0,
transaction_cost_rate=0.01,
fraction=0.5,
)
simulator = RepeatedBinarySimulator(
payoff=1.0,
loss=1.0,
fee_per_bet=0.01,
probability=0.55,
trials=1_000,
)
simulator.evaluate_strategy(strategy, bankroll)
print(f"Final bankroll: ${bankroll.total_funds:.2f}")
bankroll.plot_history(fname="bankroll-history.png")The test suite executes this exact block as a seeded, down-scaled demo — the simulator clamped to 50 trials with a fixed seed and the plot rendered headlessly — so the example stays runnable; the narrative above describes the full 1,000-trial run.
Simulation mutates the bankroll and records its history. Use matching payoff and
loss assumptions in the strategy and simulator: Keeks enforces the match — every
simulator's evaluate_strategy checks that a BaseStrategy instance's payoff
and loss agree with the odds it settles at, and raises ValueError on a
mismatch. Duck-typed strategies that carry no odds of their own are not checked —
their compatibility stays your responsibility.
Portfolios of distribution-valued options simulate through the same bankroll machinery — see the allocation documentation and the six-ETF worked example.
- Getting started — a fuller walkthrough of both layers
- Portfolio allocation — joint-return models, allocator families, and the simulator; plus the visualization helpers
- API references: strategies, bankroll, simulators
- Contract checks — runnable harnesses for extending the library
- Nine-strategy risk benchmark — what each shipped strategy actually produces under identical, seeded assumptions; regenerate every number with
uv run python benchmarks/strategy_benchmark.py - Examples:
examples/strategy_comparison.py(all nine strategies, headless),examples/st_petersburg_paradox.py(repeated sizing is not one-time pricing),examples/allocation_etfs.py(a six-ETF portfolio), andexamples/replay_bet_log.py(compare sizing rules on a recorded CSV bet log)
- Cost units differ. A strategy's
transaction_cost_rateis a per-unit fractional cost that scales with stake size; a simulator'sfee_per_betis a flat, absolute bankroll amount charged once per settled bet. The two are different units — passing the same number to both models as two different real-world costs, and Keeks does not convert between them. - Edge-aware gating.
KellyCriterionsizes on edge by default (min_probability=None): any bet its formula prices positively is placed, and a bet with no positive edge returns 0.0. Setmin_probabilityto add a longshot gate — a trial probability below it returns 0.0 even when the Kelly fraction is positive, with aUserWarningnaming the suppressed fraction. - Fractions are floored at zero and capped so the modeled loss plus transaction cost cannot exceed the current bankroll — but constraints do not prevent losses or verify your probability estimate. The models are not advice.
- [1] A New Interpretation of Information Rate - The original Kelly Criterion paper
- [2] The Kelly Criterion in Blackjack, Sports Betting, and the Stock Market - A practical guide to applying the Kelly Criterion
- [3] Fortune's Formula - The untold story of the scientific betting system that beat the casinos and Wall Street
Keeks is for educational purposes. It does not provide investment, legal, or tax advice. Models and simulations can be wrong, and financial loss is possible. You are responsible for validating your inputs and deciding whether any real-world use is appropriate.
Contributions are welcome. To set up the project and run its checks:
git clone https://github.com/wdm0006/keeks.git
cd keeks
make setup
make install-dev
make test
make lintExtending a strategy, allocator, or model? Verify it against the library's contracts before opening a PR — see Contract checks.
Keeks is available under the MIT License.