I ran a small market making experiment on Polymarket for a few months, mostly to see whether the liquidity rewards program pays enough to justify the babysitting. Two-sided quotes on a handful of markets, modest size, no directional opinion anywhere. The short version is that the rewards are real, the scoring math is more interesting than I expected, and the ways to lose money are exactly the ones you would predict if you have ever run a passive quoting strategy anywhere else. The details matter though, because the formula quietly decides which quoting styles get paid and which get ignored.
How the program scores your quotes
Polymarket adapted its rewards mechanism from the old FTX liquidity provider program. Each incentivized market has a daily reward pool, and makers split it according to a score that depends on three things. How close your orders sit to the midpoint, how much size you post, and how consistently those orders are actually live when the system samples the book.
The spread component shapes behavior the most. Every market has a maximum qualifying spread, a few cents either side of the mid, and orders outside it earn nothing. Inside it, the weighting is roughly quadratic, so a quote one cent from the mid can earn several times the score of a quote three cents out, per dollar of size. That one design choice explains why popular markets have a wall of bot orders stacked right at the touch.
Size scales your score roughly linearly once you clear a minimum, and sampling happens on the order of once a minute, with your daily score accumulating across samples. Quotes that are live half the day earn roughly half the rewards. That uptime term is easy to underestimate, and it is where the real tension in the program lives.
Two more mechanics worth knowing. The formula wants two-sided quotes, so your score keys off the weaker of your bid and ask exposure, with only limited credit for quoting one side alone. You cannot farm rewards by leaning entirely in one direction. And markets trading in the middle of the probability range are treated more generously than markets pinned near the extremes, since quoting a market that sits at three cents is close to riskless and the program has no reason to overpay for it.
What the yield realistically looks like
Every incentivized market displays its reward pool, and your cut is your score divided by the total, so the estimate is mechanical in theory. In practice the denominator is the problem. Popular markets have professional bots quoting serious size one tick from the mid around the clock, and against that competition a casual quoter posting a few hundred dollars three cents out collects a share small enough to be a rounding error.
The honest description of the return profile is bimodal. On mature, competitive markets, wide passive quoting typically annualizes somewhere in the low single digits on the capital you tie up, before counting any losses. On new markets, boosted markets, and long-dated markets nobody wants to make, the pool sometimes splits among a handful of participants, and for a while the same capital can earn several times that. Those windows close once the bots notice. Ongoing reward farming is mostly the work of finding the next thinly quoted pool before everyone else does.
One structural point people miss. Your capital at risk is the USDC resting on the book plus whatever inventory you accumulate along the way. If you are sitting on a pile of NO shares you never wanted, that is part of the denominator too, and it is the risky part.
Where you give it back
There are two loss channels and they compound each other. The first is getting picked off. When real news lands, a binary market gaps rather than drifts. Court rulings, data releases, a candidate saying something unplanned at a podium. If the true probability jumps five points, every resting order inside that jump is a gift to whoever reads the headline first, and the people racing to hit stale quotes on Polymarket are fast. The formula pays you the most for standing one cent from the mid, which is precisely where you get run over. In effect the program pays you to sell a free option on the news cycle, and the reward has to be weighed against what that option is worth.
The second channel is inventory at resolution. When your bid gets hit you own shares, and unlike a spot market maker you cannot hold indefinitely and wait to scratch out. The market resolves and your inventory goes to a dollar or to zero. Fills also cluster on the wrong side. In a trending market your bids fill and your asks sit untouched, so inventory piles up against the trend exactly when the trend is telling you something.
The uptime requirement makes both channels worse. The safe move around a scheduled catalyst is to pull your quotes and stand aside, but every sampled minute you are absent costs score, so the incentive design nudges you to stay quoted through the exact windows where expected loss per minute is highest. The makers who last treat pulled quotes around events as a cost of doing business and accept a smaller reward check. The classic blowup runs the other way. A month of grinding out rewards on a market tied to a scheduled decision, then the decision leaks early, and one stale bid gives the whole month back with interest.
Deciding whether farming beats trading
It comes down to whether you have an opinion worth anything. Reward farming is a short volatility, short information position. It pays best when nothing happens and worst when something does. Directional trading on a prediction market pays only when you know something the price does not, which is a high bar most of us clear less often than we think. If you have no edge in the underlying question, farming a quiet market is probably the better use of the capital. If you do have edge, resting bids are just discounted entries into positions you wanted anyway, and the reward check becomes a rebate on top.
My working checklist before quoting any market now looks like this.
- Compare the posted daily pool against the size already stacked near the mid. Dense bot competition means your share will not cover the effort.
- Check the calendar. Scheduled rulings, votes, and data releases are pick-off events. Prefer markets where the probability moves slowly and there is no known date on which it will move fast.
- Size positions so that a full fill on one side followed by a gap against you stings but does not change your month.
- Decide in advance when you will pull quotes and accept the lost uptime. Improvising that decision at headline time is how stale orders happen.
- Track total profit and loss, meaning rewards plus spread capture plus inventory marks. Daily reward payments feel great and are very good at hiding a slow bleed underneath them.
My experiment is still running, smaller than when it started, and only in markets where nothing is scheduled to happen for months. The payments are modest and they are honest work. I stopped thinking of it as yield in the savings account sense and started thinking of it as a retainer for occasionally taking a punch, which is a fair description of most market making. Here the punches are at least better labeled.