Monte Carlo Simulator with Unisearch random selection
Take VV out of the DARK ages with Monte
Go to cash on a certain "sell" signal (or short) then buy the top x number of stocks from a given Unisearch based on a certain "buy" signal. But obviously, the top 5 stock by RT will be different to the top 5 by CI and so on an so forth, so you do another backtest sorted by CI and other by RT etc. However, one run of a Unisearch mechanic based on picking the top (say) 5 stocks using CI or RT etc may result in a freak once off "non statistical occurance". This is crazy - all you are doing is looking at "non statistically significant" "once offs", there is a better way - A Monte Carlo Simulator with random selection generator.
QUESTION - How do we get better statistical significance in out Unisearch?
ANSWER - RANDOMIZATION and MONTE CARLO SIMULATION OF A UNISEARCH
OK let me explain
Say I have the signal to buy so I run a certain Unisearch and get a list of 20 stocks. Obviously, the top 5 by RT will be different to the top 5 by CI as previously mentioned. But lets just say they are sorted by "CI descending" and I chose to buy stocks 1 through 5 (from this Unisearch in each up/buy cycle and go to cash in each down cycle). I then backtest this Unisearch mechanic to get a result of 7% return.
However I then I run the Backtest mechanic agains but suppose I chose stock 1, 2, 3, 4 and 6 for a 105% return (a random 5 of the top 20 by CI decending). And then ran a third backtest but select stock 1, 3, 5, 8, 12 (another random 5 of the top 20 by CI) for a 185% return.
This tells me that although the first backtest appears pretty average on selecting the top 5 stocks (i.e. 7%), the Unisearch is actually pretty good as over 3 runs of selecting Random stocks from the Unisearches 20 stocks, the general return trend is 98%.
Now 3 runs is pretty meaningless. So lets say we have the POWER of a MONTGEN (MONTECARLO GENERATOR)
Say there are 7 "buy/go to cash cycles" over 3 years using your desired Unisearch Mechanic eg Using Unisearch X in buy cycles
So the 7 cycles are (1)Buy Stock - (2)Go to Cash - (3)Buy Stock - (4)Go to Cash - (5)Buy Stock - (6)Go to Cash - (7)Buy Stock
So the Montecarlo generator goes like this
(1) Buy Stocks cycle - Run Unisearch x and randomly select 5 stocks from top 20
(2) Cash cycle - Go to cash until next buy signal
(3) Buy Stocks cycle - Run Unisearch x and randomly select 5 stocks from top 20
(4) Cash cycle - Go to cash until next buy signal
(5) Buy Stocks cycle - Run Unisearch x and randomly select 5 stocks from top 20
(6) Cash cycle - Go to cash until next buy signal
(7) Buy Stocks cycle - Run Unisearch x and randomly select 5 stocks from top 20
(8) Check return over 1 run = 44% hypothetically
(9) Go back to (1) and repeat (1) to (7) 1000 times.
(10) Check return over 1000 runs = 36% hypothetically using random chance selection on a given Unisearch mechanic
You then find that overall this unisearch mechanic produces a smoothed out return of 36%. This then tells us that over a reasonable time, even if you make some errors in judgement here and there, or the market does "out of the ordinary" things, your overall Unisearch mechanic is statistically sound over 1000 runs based on unknown random elements and randomisation.
When running big $ projects that carry a lot of unknowns, you always run monte carlo simulators over your schedules to determine the most likely end date.
What we are attempting to do is use the PC processing power to smooth out the random unknowns from the market chaos. Our VV group criticises the "cherry picked" nature of the Unisearches. You can't accuse of cherry picking when you do a Monte Carlo simulator. You see, many folks don't actually buy the top 5 or 10 from a Unisearch, they generally cherry pick from the Unisearch,
If VV had a Monte Carlo Generator, it would be in a New League. The output could show scattergraphs of bunching.
WOW you guys are nuts not to do this