Why can't the market just trade in a sine wave? I hate psycho-social manipulation

You have a finite number of datapoints (trades) so of course you can fit a function that perfectly approximates those datapoints, but that doesn't mean you gain any possibility to predict further price movements.
You can also train a neural net to "predict" every datapoint in your sample perfectly but it won't necessarily make good calls for future points.
That's quite the definition of overfitting.

you both are talking about very different things

Cuz that would be too easy?

it's an academic example of an exact fit, not even done numerically but an exact solution
the term overfitting in no way applies, idiots

Why can't the market just be y = x^3

because it's exp(x)

But it has no predictive power for out of sample data and doesn't apply to any of the actual data we're talking about so yeah, it's overfitting.

>doesn't have predictive power

it's an exact fit all the way to infinity, holy shit why can't you just think outside of your cookie cutter tutorials

You said the market trades in sine waves and posted a gif where a function gets fitted to a set of datapoints (which don't even look like real market fluctuations but whatever). This fit uses a trigonometric polynomial of degree 48 for an exact solution.
Trying to explain market behaviour with a model using 97 parameters (constant term + 48 sine parameters + 48 cosine parameters) is plain stupid since fitting a model with that many values to adjust calls for failures in predictive power.

Also, from the wiki article of "Overfitting":
>In statistics, overfitting is "the production of an analysis that corresponds too closely or exactly to a particular set of data, and may therefore fail to fit additional data or predict future observations reliably"
which is exactly what was presented here and what we were talking about.

>it's an exact fit all the way to infinity

It's an exact fit on your sample data, go one step in the future and you will have a non-zero error from the real data

It's an exact fit of the data in that chart, but not an exact fit or even anywhere near a close fit of the data you are actually going to trade on.

I honestly don't think it's hard on purpose.

Yes, why not?

>make autistic joke
>sperg out when nobody gets it
it's an exact fit to infinity yeah if the data is periodic like in some DSP cases, but otherwise it's overfitting sorry. No need to sperg out.

let me guess, you would still buy high sell low.

Yes, literally any curve can be made from sine waves. There's a proof that shows this.

you mean it needs to be a real function satisfying Dirichlet conditions

It's been a while since uni, but yeah, if we're being technical... on Jow Forums.

I don't think mathematics can apply to price movement, in the sense of prediction.

Statistics can help identify tendencies (like price tending to remain around 2 std from a certain mean, in the case of the bollinger band model), or maybe stock X tends to remain 20% higher than stock Y, under certain market conditions, such as low volatility.

You can kind of formulate decisions based on what to do when there are deviations from the norm.

Price movement is solely based on supply and demand - the selling or buying into limit orders using market orders. If you're going to look for something to model mathematically, that would be were you'd be better off starting.

But getting access to institutional or retail orderflow is not always cheap or easyy to get hold of. And you wouldn't necessarily have to do anything sophisticated. Retail is generally wrong because of lack of access to insider information, and lack of volume, individually, to impact price movement.

en.wikipedia.org/wiki/Payment_for_order_flow