Market Making With A Directional Signal

The textbook market maker is directionally neutral: it doesn't try to predict whether the price goes up or down, it just profits from the spread and manages inventory. A market maker running with a short-term alpha signal breaks that neutrality on purpose — it lets a forecast of the next few seconds' or minutes' price move tilt its quotes, becoming a hybrid between a pure liquidity provider and a directional trader.

What the signal changes

Without a signal, a market maker centers its quotes on a fair value estimate and skews only in response to its own inventory — long, so lower both sides a little to encourage selling. With a signal predicting the price is about to rise, the desk skews its quotes upward before inventory even exists: it raises its bid to buy more aggressively (since it expects to be able to sell what it buys at a higher price shortly) and raises its offer to be less willing to sell into the coming move. The signal effectively adds a second force pushing on the quotes, on top of the usual inventory-driven skew.

Worked example: combining inventory skew and signal skew

A desk's baseline mid-price is $50.00 with a symmetric 4-cent spread: bid $49.98, offer $50.02. The desk is flat, so ordinarily there'd be no inventory skew. But its short-term signal currently predicts a 60% chance the price ticks up in the next few seconds rather than down — a mild but real edge. The desk shifts its quoted mid upward by a signal-implied amount, say 1 cent, producing a new center of $50.01 and quotes of $49.99 / $50.03. It is now a slightly worse deal to sell to the desk (its bid only rose a cent, less attractive to hit) and a slightly worse deal to buy from it (its offer also rose a cent, making it pricier to lift) — the desk has tilted the whole quote ladder to reflect its short-term view, not just its position. If it then also picks up inventory from a wave of buyers hitting its bid, the two skews — signal and inventory — stack, sometimes reinforcing and sometimes offsetting each other.

The explorer above shows a generic S-shaped signal curve; imagine the x-axis as "signal strength" and the y-axis as "quote skew applied" — a market maker typically maps a probability-like signal through a curve like this so that weak signals barely move the quote and strong signals shift it meaningfully, without overreacting to noise near the middle.

What this means in practice

Mixing alpha into quoting is a double-edged sword: get the signal right and the desk earns both the spread and a directional edge; get it wrong and the desk has traded away its market-neutral safety net, taking on losses from both bad fills and a wrong-way position at the same time. Firms that do this seriously validate the signal's predictive power out-of-sample and cap how much skew any single signal reading is allowed to apply, so a broken or overconfident signal can't push the book into a large uncontrolled directional bet.