[ealloc]

Summary:

-- Their trades are profitable f = 51% of the time, and they do N = 3 million trades per day.

-- Their net profitability per day is thus (well approximated by) a normal random variable with a mean of f and a standard deviation of sqrt(f(1-f)/N), or 3e-4

-- The probability of this value being less than 50% is well approximated by norm.cdf(0.5, 0.51, sqrt(f(1-f)/3e6)) which gives 2.4e-263

In other words they only expect one loss per 10^263 days, which is much larger than the age of the universe. They are actually doing much worse than expected because they lost on one day.

[pg314]

So something is obviously wrong with their model.

[phdp]

Most likely fat tails. A normal distribution is probably not a valid assumption