@mistermonster You would think so, but no, no overfittering, I used different test sets to help avoid that, and the actual principles of the algorithm is actually not that complicated. I'd be hesitant to give away the specific details as to why/how but after seeing the results I applied it to other stocks and timeframes and it consistently gave the same performance. Its a day trading algorithm and trades anywhere from once every day to a few times a day, it rides the peaks and bumps.

Made 15,000 in profit today on 150,000 in investment, whcih is better than the historical results which would have had me at around 8K a day on that level of investment. Keep in mind thats no margin use.

@freemo that's very nice, I'd be more interested in percentage of capital profit rather than dollar amounts, I'm assuming these are mock tests?

I was building a system that uses stochastic RSI and ADX to determine if something is overbought or oversold and then to determine the likelihood that the short term trend reverses. Somewhat complex but very interesting. I take an interest in what you're doing for purely academic reasons lol

@mistermonster well you can calculate one from the other.

Historic testing shows 0.8% profit **daily**... thats over 18x/1800% yearly... Real world results have been higher (partly because i pick some trades not to execute if there is a reason to speculate against it).

@freemo so you manually intervene? Still better than nothing, is there a criteria by which you manually intervene which can be automated?

@mistermonster I do, but it isnt needed. the 0.8% per day 1800% per year figures is the result you get without manual intervention. If you just let the algorithm run.. that includes a little slippage and commission fees (I overestimated both costs).

the 10% figure I saw today was with manual intervention. I had one set of stucks I let it go at completely automated (those on real world seem to get about 1.5% daily), then I use it on stocks I personally like to tell me when to enter and exit on them, so in that case it is just an assist, on those stocks with the algo as a helper it is netting me 10% so far daily (those this last fifgure will likely come down as today was an exceptionally good day for one of the stocks I keep my eye on)

@freemo is this a general purpose algorithm, like one that can be used to predict any stochastic process with some degree better than 50% accuracy over time, or is it specific to the stock market?

@mistermonster Well anythign that is truly and completely random by definition cant be predicted. so not sure any algorithm could do better than 50% on stochastic data if by that you mean full random.

So im not really sure those two statements are different things... But it was designed for the purpose of the stock market, though should be applicable to any data stream with similar characteristics,.

@freemo lol sir you're teasing me. I'm very interested for academic reasons. If you do want to discuss it at all privately I would love that. My profile has an XMPP and Matrix account if you want to talk about it with a little more depth feel free to contact me. I'm very interested in this sort of thing, for more than the ability to make money, that's cool and all but the use cases of such things beyond that are far more interesting.

@mistermonster Its a tricky place I'm in. If i explain it enough to replicate it then I risk loosing the investment advantage. Moreover if i dont explain enough to replicate it I wont be saying anything useful.

@freemo @mistermonster Get people to sign an NDA?

@2ck

Would be a must if i were to release info about it. But even then there would need to be some advantage in doing so.

@mistermonster

@freemo @mistermonster i assumed that. figure you're either looking for academic interest or considering selling (trading potential profit for reduced risk)

@2ck nope neither. It is performed so well that I cant really imagine anyone being able or willing to buy me out at a high enough price.

The risk of the algo seems extremely low. the strategy i applied has virtually no downturns over a 10 year period. and the returns are enormous.

I really cant see any way I could make more money selling it than just using it.

@freemo @2ck the reason I'm so interested is because if you've found a way to reliably predict with greater than 50% certainty whether the price of an asset will go up or down, and with no increase in risk, you've essentially found some predictable, nonrandom mover in the market. Doing so would be akin to squaring the circle, or proving P != NP, or something interesting like that. You hear something like that and it is like an itch you can't scratch.

@mistermonster

I can certainly show you the data to prove that to be true.. though its not as simple as counting number of bets wont because the algorithm bails early on bad investments and rides out good ones.. So the win to loss ratio isnt particularly telling unless you weight it by the degree of win or loss, which is what what I've done by normalizing my backtesting. The way I backtest is by ensuring that the amount invested is always a fixed cash quantity. As it goes over the amount the extra money is held as cash and not invested, if it goes below the starting amount then it invests on margin up to the starting amount. This way the profit graph shows instantaneous performance rather than overall growth (which would be exponential).

When normalized in this way it is very clear that it found a real indicator because the drawdown is extremely minimal with consistent gains on both up and down markets... one moment

Would you like me to share with you a chart of actual buy/sell points the algorithm indicated and the success/failure of it? I can show that without revealing too much actually.

@2ck

@freemo
I absolutely would love that. I'll take what I can get lol.

I'm probably going to go mad over them late into the night, because whatever math going on is going to be very interesting. Making money is probably the most boring use case of something like that. If you can pick with greater than random certainty some outcome of some scenario with whatever characteristics shared by the market you can do a lot more than make some money.
@2ck

@mistermonster

Sadly the graph wont give you any info to know the math behind them, and im sorry for that. As someone who opensources everything he does it is kinda killing me not to release it. But the money im making is just too much, at least for the moment :) Knowing me ill probably slip the algorithm into the open source world at some point and just never tell anyone it can make a fortune at the stock market and hope no one notices :) But even then not until i made my first billion :)

Anyway ill pm you shortly with an image of a day of trading and the historic performance testing results so you can see.

@2ck

@freemo
Lol I understand the hesitation, some random guy on the internet asking you to go over your philosopher's stone because he's interested isn't the most compelling reason to do it. But it isn't every day that you hear someone say they've found something truly interesting.

Just a reminder, direct messages aren't private messages, at least yet.
@2ck

@freemo lol exactly. That is why I am signaling interest while also leaving it up to you without pressure. But like i said, my interest is entirely academic and i can keep things to myself.

@mistermonster There are aspects I can share, like the profit results from historic testing, some examples of it making predictions on a chart with a limited time span... but that isnt likely to be helpful for you to get anything useful out of it.

@freemo well I am not really trying to get anything useful so much as understand what you are doing and why. I suppose that is useful on it's own. Again, if you want to talk to someone about it feel free to contact me privately any time.

@mistermonster Anything I can say in public i can say in private. If you'd like to continue discussing it, or for a link to some of the results, we can do it in public here.

@freemo yeah alright. I just figured you were trying to keep things out of public is all.

so do you use specific numerical indicators? does it rely entirely on numerical indicators (ignoring your manual interference) or does it use news sources and fundamental type stuff?

@drewfer different places depending on what im doing or using... two main places are interactive brokers and tradingview.com

@freemo are you scraping or do they have api's available?

@drewfer Well interactive brokers provides a full on API to do whatever you wish.. Trading view is a bit different. It provides charts and rendering in app, and allows you to pull up a code editor right on the chart using a special language that trading view created specifically for stock market analysis and investing. so you can write a few line of codes defining a stock investment strategy then click a button and back test it on historic data. Due to the nature of it your algorithm has access to the historic data for quite a ways back (i use 10 years back test) but you dont have direct access tot he data.

While the language is very simple, it can cover most math based analytics,and can be more than powerful enough to do everything you need there. But it does have some limitations and will only take you so far.. If you want to go past mathematical/regression type stuff it wont work. You can not and would not want to write a neural network in that for example.

@2ck Probably not in the sense I wont be shouting it from the rooftops no.. But if someone directly asked me I would be honest

@freemo 😁 i can respect that

@2ck Its not like it would be intentional. But if i loose 100K on the stock market because I was wrong somehow it isnt likely something to be exciting and make me so gitty I'd go around announcing it. But yea not intentional I wouldnt hide or lie about it.

But So far I've been using real money and its working.

@freemo @2ck how did the trading go today?

@drewfer just a few under 30K return on 200K risk

@drewfer I'm debating upping the risk to 500K tomorrow since im getting more comfortable that these results are consistent.

@drewfer I should poitn out today is not in anyway typical or representative of the algorithm. I tried a new tactic where I put the algorithm on a stock I already had faith in as a buy and hold stock (well a two stocks actually) and it wasnt on the TQQQ v. SQQQ swing today as on past days.

So part of the win was the vact that I made a really good stock pick in general, the other part is that the algorithm enhanced the profit by skipping over some bitg downturns and rebuying on the upswing.

@drewfer would you care to see a year worth of trades using the algo?

@2ck

@freemo @2ck sure!

@L29Ah I havent thrown it at any crypto yet but thats a good suggestion let me check.