Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Unsupervised machine learning with basket clusters

hackernoon.com

Post reply on HN
Print view View original

11–20 of 39 posts

Re: Unsupervised machine learning with basket clusters

#11

Post by alextheparrot »

Here's my understanding of what's going on:

He has a dataset that links companies to each element of the periodic table (Intel would likely be highly linked to silicon, while not as linked to nitrogen as a fertilizer company). He then uses some clustering and whatnot to produce groups of stocks he is trading as a unit.

How he picks what to trade is kinda hand wavy (Is he trading all baskets and only reporting the good one?). He also mentions using Google trends data, which might be some signal for his clusters. Given the lack of detail, my gut reaction is to assume he made tens of these strategies and has kindly reported the best one to us.

At the end, you see he is a financial consultant hoping to build his resume up through posts like these.

Re: Unsupervised machine learning with basket clusters

#12

Post by lightbyte »

leereeves wrote: post #5

If this approach works, is it wise to be sharing it publicly? Is there a danger than whatever gains you were able to capture with a unique insight will end up being shared with competitors?

Sharing it publicly is the author saying it does not actually work as he claims it does. You don't write a blog post on a method you created to make guaranteed money, you use it.

Re: Unsupervised machine learning with basket clusters

#13

Post by wonderwonder »

leereeves wrote: post #5

If this approach works, is it wise to be sharing it publicly? Is there a danger than whatever gains you were able to capture with a unique insight will end up being shared with competitors?

I would argue that this scenario would actually be a positive for the author. If others start using the same methodology as him on a large enough scale and trading on the same data sets they would buy when he buys driving prices up.

On the other hand when he starts selling and others sell as well speed is a critical factor as the downward pressure will make the later movers the biggest losers.

Re: Unsupervised machine learning with basket clusters

#14

Post by justadeveloper2 »

eximius wrote: post #6
justadeveloper2 wrote: post #4

Earlier quoted context omitted.

It doesn't matter if you can do it once. You have to do it consistently over years, which is not possible. I know about Renaissance Technologies, but I suspect they have insider information or some other angle and it's not all about their algorithms.

That is a rather large accusation. Can you restate that as something constructive or helpful to the conversation?

Do you understand investing at all? People have been attempting to "beat the market" for decades and nobody has been successful. There is always the "reversion to the mean" issue in every case. People with algorithms think they are beating the market then something changes and their algorithm no longer works. Except for Renaissance--they have consistently been beating the market, or so it appears.

Does that help?

Re: Unsupervised machine learning with basket clusters

#15

Post by alextheparrot »

justadeveloper2 wrote: post #14
eximius wrote: post #6

Earlier quoted context omitted.

That is a rather large accusation. Can you restate that as something constructive or helpful to the conversation?

Do you understand investing at all? People have been attempting to "beat the market" for decades and nobody has been successful. There is always the "reversion to the mean" issue in every case. People with algorithms think they are beating the market then something changes and their algorithm no longer works. Except for Renaissance--they have consistently been beating the market, or so it appears. Does that help?

He simply asked you to backup your claim, which you don't seem to be able to otherwise you would have. Throwing around a "yeah they can do it, but they must be cheating" isn't exactly constructive.

Re: Unsupervised machine learning with basket clusters

#17

Post by justadeveloper2 »

leereeves wrote: post #5

If this approach works, is it wise to be sharing it publicly? Is there a danger than whatever gains you were able to capture with a unique insight will end up being shared with competitors?

Here's how this always goes: Someone thinks they have a killer algorithm and it works and makes them money...for a time. Then it suddenly doesn't work anymore and they lose more than they gained. It is not possible to beat the market over the long term unless you are trading on insider information, somehow manipulating markets, front-running orders, or have some other edge that is not universally available. There is no Python script that will simply print money for you year in and year out it doesn't matter what data you feed it or how smart your AI is.

Re: Unsupervised machine learning with basket clusters

#18

Post by justadeveloper2 »

justadeveloper2 wrote: post #14
eximius wrote: post #6

Earlier quoted context omitted.

That is a rather large accusation. Can you restate that as something constructive or helpful to the conversation?

Do you understand investing at all? People have been attempting to "beat the market" for decades and nobody has been successful. There is always the "reversion to the mean" issue in every case. People with algorithms think they are beating the market then something changes and their algorithm no longer works. Except for Renaissance--they have consistently been beating the market, or so it appears. Does that help?

AlexTheParrot--then home come they are the only one?

Re: Unsupervised machine learning with basket clusters

#19

Post by dajohnson89 »

justadeveloper2 wrote: post #14
eximius wrote: post #6

Earlier quoted context omitted.

That is a rather large accusation. Can you restate that as something constructive or helpful to the conversation?

Do you understand investing at all? People have been attempting to "beat the market" for decades and nobody has been successful. There is always the "reversion to the mean" issue in every case. People with algorithms think they are beating the market then something changes and their algorithm no longer works. Except for Renaissance--they have consistently been beating the market, or so it appears. Does that help?

No, it doesn't help all. You have yet to provide any evidence for your accusation of insider trading (or "something else"). You can't really accuse RT of wrongdoing, just because you don't understand their methodology.

Also -- if I were a quant that managed to beat the market consistently, I would shut up and go straight to RT, for two reasons. To avoid toxic dubious comments like yours, and because RT can pay me better than your standard hedge fund.

Re: Unsupervised machine learning with basket clusters

#20

Post by justadeveloper2 »

alextheparrot wrote: post #15
justadeveloper2 wrote: post #14

Earlier quoted context omitted.

Do you understand investing at all? People have been attempting to "beat the market" for decades and nobody has been successful. There is always the "reversion to the mean" issue in every case. People with algorithms think they are beating the market then something changes and their algorithm no longer works. Except for Renaissance--they have consistently been beating the market, or so it appears. Does that help?

He simply asked you to backup your claim, which you don't seem to be able to otherwise you would have. Throwing around a "yeah they can do it, but they must be cheating" isn't exactly constructive.

Do you know the story of LTCM? If not, do not lecture me on this subject. I'm sure in time I'll be proved correct. I'm sure you've heard of Bernie Madoff and how he "beat the market" for so long, yes?

I don't have to "back up my claim" the math does it for me already! You're asking me to prove a negative--instead, show me one example of a firm that has beaten the market without resorting to funny business and I'm all ears.

Post reply on HN