Wanted to share something I'm working on relative to Pick 3. It's a bit off the beaten path, but appears to be very promising based some initial backtesting of Connecticut's Midday and Evening Pick 3. I call it BFDES, short for "Brute Force Double Exponential Smoothing".
Some background for those who are not familiar with Exponential Smoothing: per Wikipedia, Exponential Smoothing is a technique for smoothing time series data using the exponential window function. Exponential functions are used to assign exponentially decreasing weights over time, as opposed to, for example, using a Moving Average, where past observations are weighted equally. In the discussion below, I'll be using the "Holt–Winters" double exponential smoothing method, varying the Alpha and Beta variables. See Wikipedia for a more detailed discussion.
I'm in the process of running some simulations and will share the data when done (it's taking forever, due to the "Brute Force" nature of what I'm doing), but here's a synopsis of the premise of BFDES and what I'm seeing in my initial spot checks.
The premise: apply Double Exponential Smoothing to the most recent 7 draws, using Alpha and Beta values determined empirically, to identify a pool of 7 box combinations to play for the next 7 drawings. Don't ask my why this might work . . . I can't envision why smoothing recent data would provide ANY insight as to what combinations might hit in the near future . . . but my initial spot checks infer there is a benefit.
What I'm doing now is this, for one years worth of data: for a given drawing, I apply Double Exponential Smoothing to the most recent 7 draws, varying Alpha and Beta from 0.10 to 0.90 in 0.05 increments, to determine what Alpha/Beta combination yields the most box hits in the next 7 draws. After I've done this for 365 draws, I plan on digesting the data to determine if there is a tendency towards a specific Alpha/Beta combination(s) moreso than others. I would then presumably use the identified Alpha/Beta combination(s) to determine plays for the next 7 days.
So that's where I am. If anyone has any comments, ideas, buzzkills, etc, please share. I will provide an update with results after I've digested the Alpha/Beta combination(s) data, and hopefully, BFDES will eventually be refined such that this thread will move to the Lottery Systems forum!