hypersoniq's Blog

so far I have 5,000 draws calculated (a bit less than halfway).

Hard to tell the cyclic hit rate since MASS is pari-mutuel, but 7 straight hits so far (4/4) seems to be pretty good... so far...

thread is here... https://www.lotterypost.com/thread/125600

sheets available here... http://home.comcast.net/~hypersoniq

they are the full backtest versions, 5+MB each (slow d/l for dial-up)

a more compact "production" version will be made freely available when I get it done. the second version will be easier to change for other state's data.

12/29/2005

PA evening

pick3 = 0-6-2

pick4 = 2-1-1-9

after a lackluster first test (although long droughts can be expected) here goes nuttin'...

Pennsylvania Lottery 

pick3 eve 7-0-1

pick4 eve 1-6-0-1

 

12/27/2005 

pick 3 evening = 0-0-4

pick 4 evening = 8-0-5-4

finally fixed the logic...

results overall

28 years, expected 28 hits, got 64 hits!

cost to play $9,881

amount won $32,000

net profit $22,119

best year = 1980 with 10 straight hits!

years without a hit 1983, 1993, 1994, 1995, 1997, 1999, 2003

last hit 2/12/2005  2-7-0

I noticed that the 90's usually represent the downfall of most systems. I wonder why?

in all, this system is a winner, save for cyclic hit rate. SoniQ BOOM is a.k.a. the Precursor Announcer system.

I finally have something worth giving to the community, but it's not ready just yet, the sheet is COMPLEX! the process of adapting to a new state is also complex, time-consuming and computationally intense. Tonight I will adapt to PA evening big 4. IF that goes as well, I am ready to take this to the LP community for open refinement (sheet will be on my website, all 6 megs of it, when ready...)

I am not starting off with a hairbrained concept (like I have countless times in the past), I have formed a theory, and tested it to a system that passes basic backtesting... I will need fresh eyes to look at it, to pick up win opportunities I missed, to double check logic... I think this can help everyone.

anyone up to the challenge?

well, there is a bad bit of bias in the beginning, but 164 hits versus the nearest system's 27 over 18 years is a step in the RIGHT direction... all this for a single shot straight system, wonder what I can tweak to make it better, it fails on one point... cyclic hits, several periods of a few years where no hits, but some years had more than 2 or 3 straight hits...

this is a tough one, having to convert 3000 draws at a time (for each position, no less) and copy the values over the formulas to get recalc time to near zero... I am heading in the right direction, just slowly... in contrast, v-trac mirror backtest was a 10 minute affair, while I have been thinking of the SB backtest logic for DAYS until today it clicked. so far backtest results to beat for PA are 13 straight hits in v-trac (vs 11 in pure MODE()  ) best ever was 26 hits over PA history in a variation of 111, but there are too many variations... they all fail cyclic hit ratio so far.

right now I am in a recalc cycle on the pick 3 sheet that will most likely take 20 minutes, but I only have to do that once (wheew). I will convert ALL formulas to just values save for the last few rows so I can add new draws and generate new picks, but much quicker than leaving all the formulas in place.

the light came on when I realized I didn't need the count of EVERY precursor/draw pair for each drawing, just the count for the precursor/draw for that individual draw at that point in history, so I can now change system pick variables without needing to recalc the whole darn list.

I don't know how well the system will test, but if there is bias to be found, this is the system to find it...

the first backtest will be to pick the highest number of announcer occurences for the digits 0-9 for EACH precursor/draw vertical pair (will pick one number for each position, this system isolates the 3 digits and has different rolling data for each) for example, if one draw was 0-9-4 and the next was 5-2-7, looking at the first vertical pair (0-5) it will give me a count of how many times each digit (0-9) followed 0-5 vertically over history TO THAT POINT (hence the power of backtesting without hindsight--- data presented at each line is only for cumulative history, not ALL of history). IF 0-5 was followed by a 7 more times than any other number, then 7 will be picked for that position as part of the pick combo for the next game (in history, and at the end, for the next draw). When a tie occurs, I am allowing excel to deal with it and return the FIRST highest number (this will be explored in subsequent backtests).

since it works by position, it will be an easy port to the pick 4. I don't plan on using mid-day data since it is CPU drawn and I wouldn't play it anyway, so SB is just for PA evenings for now.

big 4 was a net loss, with ONE straight hit in 1995

pick 3 still pending

results = not good...

total straight wins by playing $1 straight per game in PA eve = 13 ($6,500)

total COST to play 1 $1 bet per night from 3/1/97 to 12/19/2005 = $9,874

net LOSS = -$3,374

longest gap between hits was several years  (1/1994 to 7/2000)

the hit ratio was low... now I didn't expand to count the v-trac strings, but the associated cost gets multiplied by 8, therefore no advantage...

based on my criteria, this first v-trac mirror test failed... now this was one number for one state... your results may vary, I just report what I find. Since I hold no credence in travelling numbers, I will be doing no further testing on v-tracs.

failure points

1. cost more than was won

2. cyclic hit rate way too low (big gaps of several years)

since I test straight-shooter systems only, box hits might have helped, but would also have doubled the cost.

oh well, on to the next task at hand...

pieces coming together for backtesting...

first thing about the system is there are too many numbers to play. Let me share some insights on my system design process...

To "win" at backtesting there are actually several components involved (some at the system design level)...

A). For the backtest period, amount won MUST be greater than amount played (even if only by $1)
B). Cyclic repeatability must fall within an interval of draws not to exceed the prize payout for the standard bet. (huh? o.k. for the PA lottery pick 3... $1 straight bet returns a fixed prze of $500, for a one-number system, this means it MUST not have any string of misses greater than 498 draws ((499 is break-even and 500 loses the original dollar)). In a nutshell, the system must be able to pick 1 straight winner EVERY 498 games (or less) to be declared a winner... This is a new thing for me, the missing criteria! )
C). There can be no "gray areas", it must be a system that can hold to a set of pre-defined rules that are NOT open to user interpretation in any way (e.g. the 111 system is a breeze to backtest, whereas v-tracs require far too much study (that cannot be duplicated easily if at all within excel) ) a perfect example that relates is the most frequent number... there ARE ties, a tie-breaker must be in place... for example, the MODE($A$2:$A9850) returns a single value even if ties exist... you need to be able to let excel handle the tie-breaker mechanism and take the results at face value. Mode can be autofilled down, and when there is valid data available, give a non-biased return, where a simple count() can produce ties.
D). Know your own understanding of the game at hand. This is important because you need to know what you want to begin with before you can expect excel to do it for you.

How I will be applying the above...

A) The first test period will be the entire history, other intervals will then be tested, such as a year, a quarter and a month.
B) Goal is 1 straight hit every 498 draws or less over the entire history. When I ran all the variations of 111, I found gaps of up to 5 years with no wins. Right now, exact vtrac mirrors are a winner for 2005 eve in pa (1 win), I have yet to test back to 1977 which is where most systems fail...
C) Using the unbiased output of MODE() and LARGE() will help keep from coloring the data with personal observation.
D.) This is not the "right way" to look at it, but here is how I get pick 3
---1) 3 sequential games of 1 in 10, no position has a bearing on the other
---2) IN a perfect world, past draws have no influence on future draws... What I am after is BIAS. Coincidence never hurts, but Bias is the ticket to repeatable results, and why a shorter interval than all of the history may be better.
---3) I just can't buy into travelling numbers, it's all coincidence.

On with the backtest (A bit hesitant because they ALL fail here, 3 powerball systems, 2 or 3 pick 3 systems... of my own design, not to mention TONS of other posted systems like 111 that were tested with no good result in the long run)

I have high hopes for precursor/announcer results in ALL games, if this works I might try to apply something similar to PA Cash5.

Also in the works, a multi-dimensional V-trac style experiment on powerball white ball data.

 

I probably need to work up a whole new sheet to backtest, also need to reduce the recalculation time, gotta figure that one out yet... :-(

these are for evening in PA 12/12/2005

495
493
465
463
695
693
665
663

I had to make a judgement call in the second position between the 6 and the 5... here I went to the announcer (without precursors) chart and find 6 followed 7 more times than 5 followed 7... hoping that this is a possible tie-breaking rule-of-thumb...

since this is a straight-shooter positional system, I won't be counting boxes as anything more than coincidence.

it still requires studying the numbers. gotta figure out automation and backtesting, but I have a string for tonight's PA daily number... they  are announcers, prequalified by precursors... selection criteria was most frequent, and when a tie, with most frequent in the last 3 months.

here goes nothing... PA evening for 12/11/2005...

144
149
174
179
244
249
274
279

the current announcer/precursor system on my excel "workbench" now has a name... the "SoniQ BOOM system"

here is my goal, in a nutshell...

to produce a set of 8 pick 3 combos to play based on the most frequent announcers but having the prequalified by the most frequent precursors. Basically, I am taking announcer theory and grafting it onto a binary filter system (such as v-trac strings or H/L,O/E) to extract 8 straight picks. AND, since it is locked by position, it will be an easy migration to pick4, should success warrant such a migration.

I have a list of the most frequent announcers, and the most frequent precursors. I have a history file with ALL ofPA's evening 3 digit data since 3/1/1977 for backtesting.

Several hurdles stand between idea and backtesting, but I have a ton of ideas to try. Named Ranges made it 100% easier!