3 ms·
>The numbers come from a large sample of exchanges Disclose the method and source "average price on last calendar" Average is a bad function to minimize erro
by DINKDINK 8y ago
>The numbers come from a large sample of exchanges
Disclose the method and source
"average price on last calendar"
Average is a bad function to minimize error between expected values and actual values. Use median or median of the 5%-95% percentiles.
- cryptoplot 8y agoIndeed, the median works better in general, even on degenerate distributions. The Cauchy distributions has no defined mean, but still has a median. We need to upgrade the FAQ to be more specific with the methods we use. To be quick: The price come from many exchange. We exclude suspicious points. Then we do aggregation function like the average. Averaging may not technically be the best method, but it is simple, and it has a wide support. And with the number of data points we have, we assume the Law of Large Number apply. Still it could be interesting to use the median. I need to think more about it.
- DINKDINK 8y ago>we assume the Law of Large Number apply Null hypothesis mandates you make no such assumption. Even if you did, a gaussian distribution assumes variable independence which markets are most certainly not. Even still if the variables were independent, you surely don't have enough samples of prices to report an accurate mean.
- cryptoplot 8y agoAgreed. We should not make such assumptions, because the number of exchanges is still quite low. But you would be surprised to know how many samples we have at any moment!