The main problem right now is that since the market is long-biased and there's no concept of time in the rankings, it's easy to get to the top by always clicking "Buy". It would be cool if you could make the rankings actually reflect a confidence score of how well a user is beating the market.
I'm not very strong on the statistics, but here's a possibility: for each trade, calculate the user's excess return against the market, represented by an index fund: (user's return % - market return %), and multiply that by the number of days. Then assign a confidence score using a t-test, maybe: t = (excess return * days) / (sample variance / sqrt(number of days)), and look up the confidence level based on the t-value. Finally, rank users according to their confidence level.
Something else--since this is technical analysis, it would be cool if you could include some other indicators on the chart, e.g. RSI, more than one moving average, bollinger bands, etc. That could give you some cool metrics--you could track how well each combination of indicators improves picks, by user and in aggregate.
The main problem right now is that since the market is long-biased and there's no concept of time in the rankings, it's easy to get to the top by always clicking "Buy". It would be cool if you could make the rankings actually reflect a confidence score of how well a user is beating the market.
I'm not very strong on the statistics, but here's a possibility: for each trade, calculate the user's excess return against the market, represented by an index fund: (user's return % - market return %), and multiply that by the number of days. Then assign a confidence score using a t-test, maybe: t = (excess return * days) / (sample variance / sqrt(number of days)), and look up the confidence level based on the t-value. Finally, rank users according to their confidence level.
Something else--since this is technical analysis, it would be cool if you could include some other indicators on the chart, e.g. RSI, more than one moving average, bollinger bands, etc. That could give you some cool metrics--you could track how well each combination of indicators improves picks, by user and in aggregate.