Data SourcePythPrice ConstructionOne thing to be careful with is making it robust to manipulation. If we use a single venue's 5-minute closes only, then we are vulnerable to manipulation. To prevent it, we:
1. Get the price from many credible venues, then take the median.
2. Compute 5-minute returns with 5 price sequences shifted by 1 minute, i.e., using 5-minute return sequences from prices at [00:00, 00:05, ...], [00:01, 00:06, ...], ... , then take the mean. The grid can be taken more frequently, such as every 30 seconds. When I ran a backtest myself, it didn't make much difference, though. 1 minute is already fine.Volume StrategyI think, due to its nature, it will attract a guaranteed volume from arbitrageurs and market makers. I will personally write an article arguing that this market can be a good venue for AMM LPs to hedge impermanent loss (divergence).Distribution ChannelNo