One thing that makes it hard to write a truly risk-free bot is that if I buy one asset as part of a multi-trade arbitrage, someone else may buy other assets that I was intending to buy before me, preventing me from getting the price I wanted.
This resolves YES if by the close date, there is an option in the API to place multiple bets at once. Any collection of bets made this way must either all go through, or none go through.
Real life arb bot don't have this advantage, and also I don't see why an arbbot (not run by manifold) would deserve reward for a truly risk free action.
@CamillePerrin Well ideally an arb bot would be run by Manifold. But to the question of deserts, the point of a prediction market platform is to predict events accurately as possible and they way they do that is by rewarding trades that improve those predictions, including arbs. So it seems pretty clear to me that in that sense, arbs do deserve rewards, and moreover those rewards should be good as the market can make them, and one way the rewards can be made better is by making them as riskless as possible.
@BoltonBailey Exactly. And also:
The easier it is to make risk-free arbs, the less work you have to spend dealing with the risks. Less work dealing with the risks means more bots happy to do it, which means reduced profits from doing so. Making good arb bots easy should reduce the profit that arb bots can collect.
If you push this to its extreme and allow atomic multi-bet options as limit orders then the bots can bid their own profits down to zero before someone even makes a trade that causes the arb to execute!