Live tape · simulated feed

Trade the
distribution,
not the guess.

FFB Investments runs an execution-intelligence layer that composes ten heterogeneous alpha kernels — mean reversion, Hawkes burst detection, regime-switching HMMs, policy-gradient RL — under a single risk budget you set with one dial. Every claim below is walk-forward, cost-adjusted and purged.

Ensemble Sharpe2.41
Assets live$4.18B
Median fill0.7ms
Ruin p · 10y1.9%
Run the backtest Stress with Monte Carlo
Top-of-book · regime tape --:--:--
HMM state posterior, 3-regime (calm / transition / crisis). Sizing scales with P(calm) and is vetoed when P(crisis) > 0.42.
simulated in-browser · 10s bars · no exchange data
The alpha roster · click to toggle membership

Ten kernels. One budget.
Zero correlated failure modes.

Each kernel is a separate hypothesis with its own capacity, decay curve and kill-switch. Toggle them and watch the ensemble, drawdown and turnover recompute live in the lab below.

Calibration · question 01

How aggressive is your profile?

Leverage is not a multiplier on returns — it is a multiplier on variance drag. FFB's desk book solves the geometric growth rate g = μL − ½σ²L² − c, so the dial below stops being about bravado and starts being about compounding.

VaultSystematicDirectionalConvexFFB Max
Balanced Systematic
Kelly fraction 0.34 · vol target 13.8% · 1.6× gross
Growth g+19.2%
Vol σ13.8%
Gross L1.6×
CVaR 95 · 1y−3.1%
Ruin p · 10y1.9%
Vol drag−1.5%
Within tolerance. Geometric growth is still increasing in leverage — you are left of the peak.
The lab · backtest, Monte Carlo, execution

Everything is a distribution.
So we render all of them.

CONTROLS

Engine: geometric Brownian excursion with Student-t(4) innovations, 3-state Markov volatility chain, and a cost model of spread/2 + square-root market impact + 0.3bp borrow. No lookahead, no restated survivorship.

EQUITY CURVE · NET OF COSTS

Solid = walk-forward path · dashed = equal-weight benchmark · shaded = 5–95% band
2016-01 → 2026-08

WALK-FORWARD FOLDS · OOS SHARPE

MONTHLY RETURNS · 120 MONTHS

⚠ Gross ensemble Sharpe is not the number you can bank. What survives into the curve above is Sharpe after capacity decay (edge falls off like 1/(1+(L/4.5)1.8)) and after trading costs, which are levied on turnover rather than on cleverness. The in-sample → out-of-sample replication ratio is the most honest number in quant finance: weak signals multiplied across many assets decay hardest out of sample, and more training data with fewer parameters are the only free lunches on offer.
The overfitting firewall

How many strategies did you try
before you showed me this one?

Search enough parameter space and a backtest will look like a money printer. FFB reports the Probabilistic Sharpe Ratio and the Deflated Sharpe Ratio — the second penalises you for every dead idea in your research log. Move the counter and watch a great strategy become noise.

Sharpe · net OOS1.08
Probabilistic SR0.998
Deflated SR0.981
PBO (overfit)4.1%
VerdictSurvives deflation at 12 trials — deploy at 25% size and re-test quarterly.
Paper-trading now open

Bring your hypothesis.
We'll bring the fire.

Connect a book, paste a signal, or let the roster run its own kernels against your risk budget. Two weeks, full telemetry, no capital at risk.

Request access See the numbers first