I crossed a five cent spread a while back because I was sure about my forecast, and being sure felt like it justified paying up. The contract was 37 bid, 42 offered, and my model said 45. Lifting the offer left me 3 cents of theoretical edge on a forecast that, if I am honest with myself, is rarely accurate to within 3 points. The trade could have been right on the merits and still a coin flip after execution. In a thin prediction market book the spread is often wider than your edge, and if you take liquidity by default you are donating your research to whoever posted the quote.
Prediction markets make this worse than almost any other venue. Prices are probabilities, so a nickel of spread is five full points of implied probability, and on a 40 cent contract that is more than a tenth of the price. The books are also shallow. In a mid-tier contract you might see a few hundred dollars resting at each level, which means even modest size walks the book and your average fill is worse than the screen suggested. Execution in these markets is a real part of the strategy, and treating it as an afterthought is roughly equivalent to adding a few points of error to every forecast you make.
Resting inside the spread
The basic move is simple. Instead of lifting 42, post a bid at 38 or 39, one cent above the existing best bid, and become the best bid yourself. Now one of two things happens. Either someone sells into you and you own the position at a price that preserves most of your edge, or the market drifts away and you miss the trade. People hate the second outcome far more than they should. Missing a trade costs you the edge on that one trade. Habitually crossing wide spreads costs you a slice of every trade you ever make, and it compounds in the ugly direction.
The harder question is where to rest. My rule is to post at the price where I would still be content with the fill even if the true probability had moved a couple of points against me since I formed my estimate. That sounds overly cautious until you understand why fills actually happen in thin books, which is the part of this that took me the longest to internalize.
The fills you get are the ones to worry about
A thin book can sit untouched for hours. The trades that finally happen tend to cluster around new information. A poll lands, a court filing drops, an injury report leaks, someone relevant posts something. The seller who hit your bid chose that exact moment to pay a wide spread, and people rarely pay a wide spread for no reason. So conditional on getting filled, your fair value estimate is more likely to be stale than it was a minute earlier. This is adverse selection, and it means a resting order in a prediction market behaves like a free option you have written to everyone who watches the news feed more closely than you do.
You cannot eliminate this, but you can stop being the easiest person in the market to pick off. Pull your orders before any scheduled catalyst, debates, data releases, hearings, whatever moves your particular contract. Reprice on a fixed clock rather than leaving quotes out overnight, because a stale quote is exactly what an informed seller is hunting for. And treat the speed of a fill as information in itself. If your bid gets hit within seconds of posting, do not congratulate yourself on the price. Go check what just happened, and haircut the fair value you carry for that position until you know.
Splitting size and deciding how long to wait
Showing your full size in a shallow book is a mistake for two reasons. It moves the market, since other participants reprice around a large new bid, and it advertises that someone has a view, which invites people to lean on your order. I split entries into roughly three slices. One slice a cent inside the current best bid, one slice at the best bid, and one slice two or three cents below it. The lower slices are my adverse selection discount built into the ladder. If they fill, I got size at prices that already assume some bad news arrived, which is usually exactly what happened.
How long to wait is a function of time to resolution, and I think about it as a budget. On a contract that resolves months out, information arrives slowly, the true probability drifts rather than jumps, and waiting a day or three for a fill costs very little, so resting orders are clearly correct there. On a contract resolving tonight, edge decays by the hour and a resting order is mostly a mechanism for getting picked off in the final scramble, so either take the spread with reduced size or accept that the trade is not there. As a loose rule I will wait around a tenth of the remaining time to resolution for a passive fill before I reassess, which works out to days on a monthly contract and minutes on a daily one.
The rules I actually use
- Write down your fair value before you open the book, so the quote cannot anchor your estimate.
- Compute your edge at the ask. If it is smaller than your typical forecast error, taking liquidity is gambling on execution, so rest or pass.
- If the edge at the ask is genuinely fat, just take it. Being cute with a limit order to save one cent while risking a ten cent move away from you is bad arithmetic.
- Rest at the price where a fill would still be acceptable after modest bad news, and ladder the remainder of your size below it.
- Never leave a resting order live through a scheduled event. Cancel first, reassess after.
- Reprice or cancel on a fixed interval. An order you have not repriced in hours is quietly working for the other side.
- When a fill comes back fast, check the news before you add to the position.
None of this rescues a bad forecast, and I want to be careful about overselling the mechanics. What it does is stop decent forecasts from dying at the point of execution, which in illiquid contracts is where I suspect most of them actually die. My better stretches in these markets have come from ordinary opinions executed patiently, and the losing streaks, when I go back through the fills, usually trace to a week where I got impatient with the book and started paying the spread just to feel involved. The forecast gets the credit either way, so it took me an embarrassingly long time to notice that the real difference was execution.