How Ballast is built

The idea is old and well tested: hold assets that earn in different economic weather, size them so each contributes a similar amount of risk, and rebalance on a schedule instead of on a hunch. What is new is running it in the open, with an AI operator and a public record.

The all-weather thesis

Most portfolios are quietly a bet on one regime. A stock-heavy book does well while growth is strong and gets punished when it is not. Ballast starts from a different question: what do you hold so that something in the book is working no matter which way the economy breaks?

There are really four weathers, and each has an owner in the book.

The sleeves

Each holding is tagged to a sleeve, and each sleeve has a job. This is what earns Ballast its keep across regimes.

The aggressive book adds the crypto kicker; the balanced and conservative books leave it out.

Risk-parity sizing

Weights do not come from a view on what will go up. They come from volatility. Each sleeve is sized so it contributes a comparable share of the book's risk, which means the quiet holdings get more dollars and the wild ones get fewer. A near-riskless holding like T-bills would otherwise swallow the whole book under this math, so its volatility is floored before weighting. The result is then scaled toward the book's volatility target, and a hard leverage cap sits on top of everything.

That last cap is not a suggestion. Share counts are always floored, never rounded up, so the book cannot drift past its leverage ceiling at construction or after any rebalance.

The three risk levels

Values are set in each book's mandate and are the single source of truth for the engine.

Rebalance discipline

The book is rebalanced on a monthly cadence, and also if a sleeve drifts far enough from its target weight to breach the mandate's drift band. When that happens the engine recomputes the target through the exact same sizing pipeline used at construction and produces a trade list. Net asset value is preserved across a rebalance: it is a reallocation, not a deposit or withdrawal. Drift-triggered moves are published as part of the regular letter cycle rather than fired off as ad-hoc alerts, which keeps the cadence steady and the posture clean.

Ballast is paper only. A rebalance produces a trade list as data. There is no order surface and nothing executes.

The risk engine

On top of the sizing pipeline sits a small risk engine with two levers. An exponentially weighted volatility forecast reacts to regime shifts faster than the flat lookback the sizer uses, and a drawdown circuit breaker scales exposure down along a ramp between each mandate's soft-alert and hard-review lines, then re-risks automatically as the book recovers. The design rule that matters is what it refuses to touch: the engine only ever trims the leverage above fully invested, because a book that borrows nothing has no leverage-amplified risk to cut, and cutting it anyway just locks in an ordinary loss. On the unlevered books this makes the engine a provable no-op, byte-identical backtests either way, while the levered book keeps a genuine circuit breaker over the borrowing it actually does.

How much of this is just equity beta?

The first question a serious allocator asks is how much of a return is compensated market risk rather than genuine skill. Ballast answers it directly. Each book's backtested daily return series is regressed on the standard academic factor set, the Fama and French five factors plus Carhart momentum, sourced free from the Ken French Data Library. The output reports the factor betas, the equity-market beta chief among them, and splits the return into a factor-attributed part and a residual. The residual is the honest measure of edge beyond known factors, and it is the number we lead with.

The decomposition is run on the backtest series, not realized-forward performance, because the live books have almost no forward record yet, and it is labeled that way everywhere. When a book turns out to be mostly equity beta with a small residual, that is reported plainly rather than buried. Here are the current committed figures:

Backtested daily returns regressed on the Fama and French five factors plus momentum, from the Ken French Data Library. Classical OLS standard errors, so treat the t statistics as an upper bound on significance; nothing here has been blessed by the deflation gate.

How big could this run? The capacity ceiling

A backtest that ignores its own market impact looks identical at any size, which is exactly how funds end up selling capacity they do not have. Ballast's backtests charge a square-root market-impact cost on every trade, and from the same model we solve for the assets under management at which the strategy would start to eat itself. The rule is conservative: no rebalance trade may exceed a tenth of an instrument's daily dollar volume, and impact may never consume more than a quarter of a sleeve's gross edge.

Every book's ceiling binds on participation, meaning the wall is how much the least liquid instrument actually trades in a day, not the erosion of some fragile edge. That is what makes the number fundable rather than flattering: it comes from public volume data and a stated rule, so you can recompute it and argue with the rule instead of trusting us. At the fund's current paper size the impact cost is immaterial, which is precisely what the model should say about a small book in deep markets.

Reverse stress: what would break each book

An ordinary stress test asks how the book does in a chosen scenario. The reverse question is sharper: for each book, what is the smallest version of a bad day that drags it all the way to the drawdown line where its own mandate demands a hard review? We keep a library of forward-looking joint shocks (a parallel rate shock, a breakdown of the diversification the book is built on, a volatility spike with a momentum unwind, and an inflation shock) and solve in closed form for the scale of each that reaches the line. The scenario needing the smallest scale is the book's soft spot, named below.

Computed from each book's current sleeve exposures and its own mandate's hard-review drawdown, using the documented shock assumptions in the scenario library. A scenario whose direction nets positive for a book cannot be scaled into a loss and is reported that way, never forced.