Quant strategy

Depends. What's the PNL? Are your factors actually correlated with the performance of the assets you've chosen? What creates your signals? What is the underlying idea for each part of your strategy?

So many questions need answers for me to provide a reasonable response.

There you see it - you already have a strategy in mind, you're now analysing 200 pieces of data with the intent of fleshing out that strategy/ hypothesis. That's data mining - you can always find evidence within a window of data for certain strategies as long as you keep mining the data and trying out diff methodologies of sampling. Not to mention 200 isn't long enough. I'm no expert in quant strategies but isn't there barely any evidence for mean reversion and TA? Some food for thought

50%, same probability as if u toss a coin

As another one said it is clearly data mining.

There are a ton of things to check when doing backtest, too much to explain by writing

Its probably easier if you try to get an entry level job at a reputable quant fund to learn the ropes first. If you try to each yourself, you're probably going to data mining as others have mentioned- ex: come up with a fantastic backtest that doesn't work live. Because if you try enough different things, eventually one of them will work by random chance. One way you can help yourself is by using in and out-of-sample- keep some part time period of the data (say the last X number of years) completely separate and untouched, and then don't look at it at all when developing your strategy. Then, at the end, when your strategy is finalized (and won't change anymore), evaluate how it would have done on this separate period of data. If it works on this separate sample, that is a better indication it will work live (but not a guarantee).

Lot of good info here. Unfortunately, like many other commenters state; I do think that you need to have a little bit of humble pie before you go looking for investors to fund your portfolio.

An alpha that predicts/ uses (only market/ price?) data on only the top n winners and bottom n losers sounds like a wild strategy for small n (and probably large n). Maybe your alpha is more event driven and greatly depends on monetizing earnings events or other things of that ilk, idk, you’re 17 and that probably sounds like gibberish to you.

Okay, so if you ended up creating a L/S portfolio off whatever in gods name free data you’re using to predict top winners and losers (? Or only just use that for universe selection, still makes me queasy), you’re going to have to tell me a lot about your portfolio construction process for me to discern whether you’re really just betting on factors or not (another sentence that probably makes no sense to you). In short, an example would be that you could gain access to the returns of something that tracked the market, even levered or “outperforming” the market as you state (big ups yourself that you’re using a benchmark), say even several X times the cap weighted market portfolio return. That exists, is cheap (in some sense ie the fee there relative to paying an active manager), and no investor is going to give you money to go and do that. Additionally as long as they’re not idiots it’s pretty quick to tell whether you’re just riding factors or not (especially with position level data, but for an investor they’ll could probably do it with just correlations of portfolio level returns). Now there is some theory on factor betting and edges there/ what even defines a “factor” in a philosophical sense but long story short you may have just developed a model that works on predicting something that is notoriously hard to predict (just by its own price/ returns data) that isn’t really differentiated wrt actually having “idiosyncratic” alpha/ returns.

Moving off that topic I will echo some of the other responders here along with my own institutional stat arb pm insight of: y’a need to buy the data to extract the alpha to make the portfolios to then go and systematically trade (another difficulty there but one commenter mentioned paper trading and that’s a good idea to start). Maybe just maybe if you have great feature selection/ modeling/ experience you could get away with 5 (very different) datasets to create you’re predictions but that’s still gonna cost more than you have to spend, (like hundreds of thousands/ year). Also in terms of that 60% right number the above will mayyyybe get you a 10bps edge daily, longer term/ frequencies of course have bigger returns but ofc also the farther out the harder to predict (with skill)! Mkt data itself, even if fully tapped out wrt realizing potential predictive power, just isn’t going to be enough. Maybe you could tell me more about your data/ process for me to understand what’s going on. Terms you need to look up there are breadth, IC, and IR. What your describing sounds like clairvoyancy or just memorizing the data/ having undisciplined process.

TLDR;

Now is the endeavor completely useless? No, it’s a learning experience and will have you run head on into understanding various terms, phenomena, and methods wrt managing a portfolio that will help you figure out what you want to do well before other folks of your age have. You mentioned a lot of data/ quant/other things. I’d recommend Grinold and Kahn’s Active Portfolio Management as a place to start/ get some theoretical underpinnings. You’ll need to know linear algebra (most 17 yo don’t) at a working level as well as some stats/ math that a 17yo has probably started to run into (basic stats and probability). Maybe you’re starting to see the picture here, becoming a quant, designing and implementing robust statistical insights/ optimizations/ modeling and more is something that takes a lot more training to do competently in any sense. Good luck on your journey.

Side note someone mentioned TA and mean reversion don’t really have alpha, my two cents: traditional TA, nope, none. Most optimal learning / policy/ whatever you do for alpha combination quickly shoots TA alpha weights to zero in my experience. “Mean reversion” - literally the simple stuff techniques, not much, but like mean reversion as a principle/ for signals/ features is fine/ used all the time, whether it works or not is just dependent on the data itself, ie maybe using analyst estimates data with some reversion principles does show mean reversion tendencies, or option implied vol metrics might etc etc. But yeah like simple historical price MACD stuff, not so much