Take that set of numbers, including the zeroes, and calculate the RMS. (square them all, sum the squares, take the square root of the sum).
...
This downside deviation method is pretty similar to DDDD3 isn't it? I think DDDD3 used a fixed target.

Same idea, just a subset for simplicity.

The DDD3 uses both rolling quarter and rolling year figures, the 3 meaning a triple weight on the rolling year figures versus only single weight on the quarterly.
So this metric is basically 3/4 the same as DDD3.
I included the rolling quarter figures in the definition DDD3 to minimize emotional pain, not really to create the best portfolios--it's the longer term that really matters : )

DDDD3 is just the daily version of DDD3. The extra D is for "daily".

With daily calculation you get 252 hold period start dates per year instead of 12, but of course it's still only one year of real world results--the overlaps are so significant that it isn't really much more in the way of sample size. Going daily helps more in a backtest than when measuring the results of a real world portfolio, because in a backtest you could in theory have different picks every single day. In a real world portfolio your picks aren't changing during a hold period. (my portfolio is running a two month cycle).

All of the above are just particular instances of downside deviation, as written up by Mr Sortino. In essence, it's a measure of the probability that any one year interval of the portfolio will fail to make the return hurdle (with a squared penalty on the size of any shortfalls below the hurdle).

Side note: Being at heart a probability that something will happen in a time interval of a particular length, it can't meaningfully be annualized, since equity portfolio returns are not random walks as time frames get longer. In the general case, an options strategy might have a high probability of a given sized loss at the one month mark but a low probability at the one year interval: a slowly rising but jagged line. Annualizing the biggish monthly number would instead give you a huge annual number equating to a virtual certainty of failure, the exact wrong result.

Mr Sortino's key insight in his creation is that a positive return below your bare minimum is still a failure. If you need 8%/year to avoid eating dog food at age 80, or if you're a pension fund that needs 8%/year to make your payment commitments, then achieving only 7% every year is a true risk. But you have to remember that the threshold to choose for the calculation is NOT the number you want, it's the number you need: the number below which your portfolio has truly failed.

I picked a pretty aggressive 10% only because this is in the context of quant screens, where backtests are always way higher return than the resulting reality. So the original 10%/year minimum threshold I picked could be thought of as (say) 6%/year of actual return and 4%/year of exaggerated optimism in the backtest : )

Jim