Stock market prediction stats

[NoDaddyNo]

I finally have enough data that I have tracked and I can give some real stats on what my 3 models do (I have more models, but these are the ones that are the base for now - the rest are still being tweaked and accumulating proper data).

I am going to add in analysis per day/week/month as well to see the fine grain (currently it just tracks everything since the day they were activated).
If something is "dead" then it means the stock changed names, or exchanges, or just went under - they don't always get pulled out of the database in time, or they go under in the time that the algorithm watches.
Bad Data is when for a dataset that I am looking at, I don't have data for it - I have a shitty data feed right now, so that happens a lot - will never happen once I upgrade to a good feed.

These have run for a bit, but they have only been tracking for a little over a month. That means that it is through earnings - which are notorious for being ass-raping times.

I'm later (hopefully today) going to add in a random selection of stocks that match the same amount of high vol/low vol stocks and see how they perform - just to show what random selection would give you.
Then after that, I will show paper trades with a range of different amounts of money on hand and different trading fees.

Each day that is represented below is a trading day - so for the first one, it is 2% gain over 21 trading days - which is an average month of trading days.



See below...


2 percent over 21 days
Count RawTotal: 1116
Count TotalMinusDead: 1066
Count Right: 983
Count Wrong: 83
Dead Tickers: 50
Percent Right: 92.21
--------------------
-Volume based-------
--------------------
High Vol Percent Right: 93.75
Low Vol Percent Right: 91.35


1 percent over 5 days
Count RawTotal: 896
Count TotalMinusDead: 847
Count Right: 569
Count Wrong: 278
Dead Tickers: 49
Percent Right: 67.18
--------------------
-Volume based-------
--------------------
High Vol Percent Right: 70.74
Low Vol Percent Right: 64.65


.5 percent over 1 day
Count RawTotal: 823
Count TotalMinusDead: 785
Count TotalMinusBadData: 708
Count Right: 416
Count Wrong: 292
Dead Tickers: 38
Bad Data: 77
Percent Right: 58.76
--------------------
-Volume based-------
--------------------
High Vol Percent Right: 66.41
Low Vol Percent Right: 54.26

 

[WODIN]

I wanna fuck you till you bleed! Cool stuff dude.

 

[Lumberg]

Looks like you're better at the long term stuff. However taking for example a year time frame half a percent compounded 250 times adds up to a lot more than 2 percent compunded 22 times!

Lets see 1.005 to the 250th power.... 3.48...lets say triple your money to be safe


1.02 to the 22nd power 1.54...lets say 50 percent annual return

Obviously the forst on is a lot more work, and this is a very rough estimate not including off days with negative return.....but not bad

BTW there are about 20 trading days in a year right?

 

[FatRat]

Nice work Martha. I expected the 0.5 to be higher judging by the rersults of the ones you posted, but I guess they are subject to a lot of random fluctuation, particularly over reporting season like you said. The 2% looks promising!

Did you ever get anywhere with the london data?

 

[NoDaddyNo]

Nice work Martha. I expected the 0.5 to be higher judging by the rersults of the ones you posted, but I guess they are subject to a lot of random fluctuation, particularly over reporting season like you said. The 2% looks promising!

Did you ever get anywhere with the london data?

That .5% one seems to be the most susceptible to the earnings data.
Also, the .5% one is a bit of a hack - the code that is doing the prediction is testing for one thing, but then the trading method on it is doing something else.
If it were setup correctly, so that the program were actually optimising the proper thing, then it would perform a lot better.

But I mainly was using a single day so as to show that the thing did fairly well even outside of its programming ability, and also it was easy to track day to day and not have to say "check it out in a month".

It will show better when it compares to more elements in the stats - those are just a few I banged out last night.

It needs to show the random selection, the real trading performance (well, paper trades with fees), and then also comparison to index returns over the same period.

I have yet to add in about 25 tweaks that should improve these tremendously - I hadn't so far since 90%+ on something that performs pretty quickly is pretty good - now I will optimise it for performance and it will probably take longer to compute, but it will be even better at predicting.

I have yet to add any foreign markets to the data - there are a few that are staged to go into the database - but in order for those to be in, I would have to rewrite a lot of code so that it could be aware of all the data, or just specific markets. Currently the code just chomps through all the USA data.
I thought it would be an easy fix to change that, but the amount of code that would have to change is actually pretty sizeable - so I think I'm better off just adding in a whole new database of foreign markets (one database for each).
And then write code to grab from N databases, where N is what the client desires.

All of that is still months out.

My site design is 95% done - it needs a bit of code in the backend to allow automated payment and then I just have to wait on the card processing and the bank deposit system to be setup (that is largely out of my hands and depends on who I use for that).

Initially I was going to try to do it tax free here in Bermuda, but the laws are sketchy enough that it isn't clear to me what is allowed - so I'm going to do it all in the States and then if it makes enough money where I start to care about it, then I will hire a tax attorney and see what my legal options are for it.

 

[gonelifting]

Yeah that 2% looks great, but what did the overall market do in that same time period?

 

[NoDaddyNo]

BTW there are about 20 trading days in a year right?

??
I assume you mean a month?

 

[NoDaddyNo]

Yeah that 2% looks great, but what did the overall market do in that same time period?

Exactly.
That is why I need to add in more stats and show the indexes and how they perform, etc.

Also, note that these are in no way performance figures yet - these are prediction figures. So it means that it was right N% of the time - the amount potentially made on it isn't up yet.

I can tell you right now that if it is predicting 90%+ that it is doing its job since I have been running this for a bunch of months now (very bad months) and I have seen the same thing.
I just hadn't been tracking the stats in an automated system like I am now.

Incidentally, fund managers love anything that gives you the edge - so even if something is consistantly predicting 52% is just fine by them.
That is why I'm going to show the random "predictions" with it.

 

[FatRat]

Yeah that 2% looks great, but what did the overall market do in that same time period?

Good point, its been a very good month...

 

[FatRat]

Cool Martha, how did you get on with the interested backers?