Is bad algo always bad?

[HarrySquatter]

Hello!
The question is purely theorethical. Let's say that I have developed an algorithm that is predicting whether the market will close lower or higher. If my algo is 80% wrong, is it a way to just trade against its predictions? Or its very rare and the most of the time the algos are very close to the coin flip when it comes to predicting? I'm just starting with automated trading, so please show understanding and correct me if some of my assumptions are wrong

 

[vanzandt]

Hello!
The question is purely theorethical. Let's say that I have developed an algorithm that is predicting whether the market will close lower or higher. If my algo is 80% wrong, is it a way to just trade against its predictions? Or its very rare and the most of the time the algos are very close to the coin flip when it comes to predicting? I'm just starting with automated trading, so please show understanding and correct me if some of my assumptions are wrong

I have no idea... but if you're already at 80%... if you can just get that algo to be wrong 100% of the time... you'll be a billionaire.

 

[Xela]

If my algo is 80% wrong, is it a way to just trade against its predictions?

An algorithm being wrong 80% of the time is the same (as you infer) for directional purposes as its being right 80% of the time. "Right" and "wrong" are in the eye of the beholder: all you need to do is include a "reversing instruction" to transform one into the other. Algorithms don't understand "right" and "wrong": they just comply with their instructions.

The question is therefore simply one of whether you can construct a profitable trading system from a directional indicator that's "right" 80% of the time. And the answer to that question depends on a large range of variables and parameters not mentioned in your question.

The superficial or intuitive appearance (and the easy mistake to make) is the assumption that "it 'must' be".

The reality is that it may be, but it's probably going to take a lot of painstaking, methodical research and testing (and of a lot of different parameters) to find out.

I don't mean this as a criticism in any way, but in general, the question you're asking (which is of a type comparatively often asked in trading forums) tends to arise from a place of "thinking about the acquisition of trading skills primarily in terms of getting the direction right more often than wrong". Call me a skepchick, but that's not actually a frame of reference I found very productive, myself, when I was starting off. (I cast absolutely no aspersions at all, but it's one that's also often related to some confusion between "trading systems" and "entry methods".)

And welcome to ET.

 

[zdreg]

An algorithm being wrong 80% of the time is the same (as you infer) for directional purposes as its being right 80% of the time. "Right" and "wrong" are in the eye of the beholder:

And welcome to ET.

I always thought that it is beauty that is in the eye of the beholder. then again I can see the beauty of this algorithm.

 

[Error Correction Funder]

It's more likely that you'll suffer from sampling error or outright bad modeling.

 

[HarrySquatter]

An algorithm

...

And welcome to ET.

Thank you for your explanation. When I don't understand, I ask. When I ask, the question may be poorly formulated as well. But that's the way I learn and your answer helped me get some intuition, that will lead to more accurate questions in future.

 

[d08]

Hello!
The question is purely theorethical. Let's say that I have developed an algorithm that is predicting whether the market will close lower or higher. If my algo is 80% wrong, is it a way to just trade against its predictions? Or its very rare and the most of the time the algos are very close to the coin flip when it comes to predicting? I'm just starting with automated trading, so please show understanding and correct me if some of my assumptions are wrong

Slippage and commissions. Also, 80% says nothing about the sizes of profits and losses, which is more important.

 

[tommcginnis]

Slippage and commissions. Also, 80% says nothing about the sizes of profits and losses, which is more important.

Your question is an excellent one, Harry Squatter, and too often, the lesson (and the answer) are forgotten.

The lesson is, "Summary statistics" (even something as basic as "profit/hit percentage" does not tell the whole tale.

The answer is, that your misses (so often) tell the more important tale of how to improve the method (the algo).

Consider Anscombe's Quartet [four sets of x,y pairs with the same mean and variance] -- which should be a *first*principles* lesson in finance, but often gets pushed off to some preliminary stat course, where it is labeled a curiosity and promptly forgotten. In any event, your "80% miss" take-away does not reverse itself to any sort of "80% hit" rate, should you reverse the trade. Nope nope noppers nein.
https://en.wikipedia.org/wiki/Anscombe's_quartet

So, "Great Question!"
And the answer?? The *reason* for the miss is more important than the percentage of the miss.

 

[userque]

Hello!
The question is purely theorethical. Let's say that I have developed an algorithm that is predicting whether the market will close lower or higher. If my algo is 80% wrong, is it a way to just trade against its predictions?

Hi!

Yes. Trade opposite the predictions.

Or ...

Your question is flawed in that these two circumstances aren't mutually exclusive. IOW, it's like me asking you, "is it raining, or snowing outside at your location right now?"

...its very rare ...

"It" depends.

...and...

Again, as to structure of your question. You're (again) presupposing an answer to your own question: that each condition on both sides of your "and" are both true for the purposes of your question.

...the most of the time the algos are very close to the coin flip when it comes to predicting?

It depends. Notwithstanding the fact that I don't have data on "most" algos, I would nevertheless guess this is true.

I'm just starting with automated trading, so please show understanding and correct me if some of my assumptions are wrong

I'm curious as to why you're curious.

 

[eves_banished_children]

If it's 80% wrong on a binary choice.