PancakeSwap has a price prediction game(https://pancakeswap.finance/prediction), and I notice that the payout of Bull and Bear often differs a lot. Therefore, there might be a small arbitrage opportunity. This bot bets and claims rewards automatically. Data shows that the mean return is positive, usually above 15%, based on the config. However, our assumption that the win rate is 50% for all the bets is incorrect. Evidence shows that the higher the extra return, the lower the possibility of winning, which means that participants can predict the market to some extent.
The distribution of EV ex-post (the EV that takes effect after betting instead of the expected EV before betting) tends to look like this:
The predicted win rate for bets with abnormal returns, using a logit regression model, which is quite close to the historical win rate:
After the adjustment for the reduced win rate for high returns, the arbitrage strategy cannot produce extra return. The initial assumption that compelled me to do this project is wrong, but why could people predict the price of BNB in 5 mins? Shouldn't it be impossible based on the random walk theory? This phenomenon requires more research.
You need a BSC address, its private key, and an RPC link for this bot. If you have prepared these items, put them in the key.py file accordingly. If you don't have an RPC link, you might use a public RPC listed here https://docs.bscscan.com/misc-tools-and-utilities/public-rpc-nodes. You also need to download the library web3 before running main.py. You should set your preferred EV and bet amount parameters.
This is a project for education purposes. The chance of winning money by running this bot is minimal.