Copied from the discord suggestion:
The current loan system has a bunch of great effects, including making long term markets more viable, enabling people to interact with the site when already invested, enabling leverage for diversified positions, and more. But it also has some real drawbacks, especially including people defaulting on loans, people being unable to exit underwater positions without paying back the loans and going negative, and incentivizing cashing out profit because of opportunity costs. It also means that there is a huge liquidity event (positive or negative) any time a large number of markets resolve (or even one important one) and unrealized profits and losses get realized. Available balances may suddenly experience a huge jump, even if the market had been very stable near its eventual resolution value for a long time. The current system handles loans per-market, after a global cap for no loans on underwater accounts. The loan is
0.04 * max(0, min(cost basis, investment value) - existing loan amount)evaluated per market and totalized. If your loan is underwater you stop getting loans but don't have to pay it back; if you make a profit you only get it on the cost basis and not the profit. My proposal:
max loan amount = 0.05 * account net worth
loan amount = 0.02 * (
max(0, min(cost basis, investment value, max loan amount) - existing loan amount) +
(min(investment value, max loan amount) - existing loan amount))Evaluated per market, with negative amounts clawed back from the account balance. That is: you can't have a loan > 5% of your net worth on one market. Half the loan amount is calculated the like the current way (but capped at 5% of net worth per market); half is purely on the investment value, which will both loan back profits and claw back underwater amounts (but more slowly than the current system). This is really two separate changes, but I think they go naturally enough together to be worth discussing together.
Half the bounty for a PR that I think looks correct and implements the math the way I want. The other half split with author and a competent reviewer who agrees it looks good and/or helps fix it up until it does. (My js reading ability is limited, I do not know the Manifold code base, and I'm not going to pretend I can provide competent review.)
The exact constants here don't matter and are presumably configurable. Pull request being merged is not required. Ideally we'd get technical feedback from the Manifold staff, and the latter half of the bounty would require them to have no technical objections. If the PR is sound but they don't want it as a matter of policy this bounty will still pay out in full.
Exact config value for the max loan vs net worth is entirely unimportant to this bounty as long as it exists as an option (and even that is up for debate, especially if it's replaced with something that serves the same goal). I currently think it should probably be as high as 100% or so. That catches the worst problems without being a drastic plan. Adjustments can come later.
Several large users have positions on single markets as high as 80%+ of net worth, and forcing that to change would be more drastic than this market is intended to be. That sort of question is a risk management question for Manifold; this bounty is just trying to make sure that tool exists.
Fixing (at least in part) the weird incentives around unrealized profit / loss is the more important half.