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

Using a Keras Long Short-Term Memory Model to Predict Stock Prices

heartbeat.fritz.ai

Post reply on HN
Print view View original

91–100 of 178 posts

Re: Using a Keras Long Short-Term Memory Model to Predict Stock Prices

#91

Post by ttobbaybbob »

throwawaymath wrote: post #81
DevX101 wrote: post #26

I'm not a purist believer in the efficient market hypothesis. BUT, I doubt there's much alpha to be gained simply from looking at price data which is widely and publicly available. Also keep in mind that markets are dynamic feedback loops so even if this model had an edge, the act of publishing this article would work against you to neutralize that edge in the future. There's a good reason the most successful 'quanti…

I'm pretty sure that lifetime NDA is a myth. It gets repeated very often (like a lot of the Renaissance mythos) but I haven't seen evidence for it, and on its face it sounds flagrantly unenforceable. It could be that the compensation alone substantially reduces turnover. Likewise once they've hit "their number", employees may decide to go into tech or philanthropy instead of a competing firm. For what it's worth, Ren…

The NDA/non-compete may not enforceable, but if Renaissance believes employees with valuable IP are being poached, they will sue and make it very expensive to hire, then defend, ex-rentec employees.

https://www.marketwatch.com/story/renaissance-millennium-set...

Re: Using a Keras Long Short-Term Memory Model to Predict Stock Prices

#92

Post by turingcompeteme »

SEJeff wrote: post #75
turingcompeteme wrote: post #65

Earlier quoted context omitted.

Bigger sure, but can you point to one that has had better performance than the medallion fund?

Most of them aren't hedge funds, they're private (HFT, Electronic Trading, Botique Firms, Proprietary Trading are all terms you'll hear) so they don't have public data. It is literally impossible to prove without insider information. However the Virtu Financials, the Citadels, etc, will always exist, and will be doing exceptionally well whether people realize it or not.

I realize this doesn't actually matter and I feel weird defending Renaissance... but Virtu is a public company and hasn't been doing that well, and returns from Citadel's various funds are not hard to find. Also, I don't know how to compare a fund's returns to that of a private business. But Medallion has been around since before HFT was really a thing, and I'm not aware of any HFT places that have been growing at 70% a year for 20 years.

Re: Using a Keras Long Short-Term Memory Model to Predict Stock Prices

#93

Post by goodside »

This is pure nonsense. This isn’t even the right way to begin thinking about this as a forecasting task — the target series should be log-normal returns, not raw asset price. The performance of this model is laughably bad, which is probably why he spends zero time evaluating its effectiveness. You could trivially get better forecasts than this by naively repeating the last-observed price.

This isn’t ML. It’s cargo-cult performance of words and ideas that ML people use.

Re: Using a Keras Long Short-Term Memory Model to Predict Stock Prices

#94

Post by currymj »

nabla9 wrote: post #78

The author uses stock market as a toy example. He is not actually attempting to do serious forecasting. Only comparing the prediction accuracy against simpler models like ARIMA or VARMAX (as the author suggest) can tell you if the model has any use.

It seems like if you do anything with financial data like this, people often aren’t capable of accepting that it’s just a toy example — they judge it as if it must be intended to really make money. I don’t know why.

Re: Using a Keras Long Short-Term Memory Model to Predict Stock Prices

#95

Post by T-A »

DevX101 wrote: post #26

I'm not a purist believer in the efficient market hypothesis. BUT, I doubt there's much alpha to be gained simply from looking at price data which is widely and publicly available. Also keep in mind that markets are dynamic feedback loops so even if this model had an edge, the act of publishing this article would work against you to neutralize that edge in the future. There's a good reason the most successful 'quanti…

"A research scientist and senior level employee who worked out of his Pennsylvania home […] His noncompetition agreement prevented Magerman from working for one year after leaving Renaissance for any firm engaged in the business of mathematically-based trading of futures and securities." [1]

Not quite lifetime, unless they expected him to drop dead within a year.

[1] https://www.forbes.com/sites/nathanvardi/2017/05/08/inside-t...

Re: Using a Keras Long Short-Term Memory Model to Predict Stock Prices

#96

Post by woah »

lifeformed wrote: post #77
C1sc0cat wrote: post #50

Earlier quoted context omitted.

A lifetime noncompete really I assume they pay you for life for this noncompete.

Well, being an employee gives you exclusive access to the best fund in existence by far (35% return every year for decades). Nobody is going to want to kill their golden goose.

Extreme secrecy plus consistent very high returns sounds like the hallmarks of a ponzi scheme

Re: Using a Keras Long Short-Term Memory Model to Predict Stock Prices

#97

Post by Symmetry »

everdev wrote: post #35
DevX101 wrote: post #26

I'm not a purist believer in the efficient market hypothesis. BUT, I doubt there's much alpha to be gained simply from looking at price data which is widely and publicly available. Also keep in mind that markets are dynamic feedback loops so even if this model had an edge, the act of publishing this article would work against you to neutralize that edge in the future. There's a good reason the most successful 'quanti…

These predictions can also be self-reenforcing. If enough money believes the model is accurate, it will create it's own market conditions. In a distopian AI stock prediction world the best model will be the one with the widest publicity and adoption, not the best data points.

Generally when talking about price movements the opposite is true. That is a prediction that the market will move at some point in the future will cause that move to happen right now instead, but only if that prediction is believed.

And if you have a model that people believe which predicts the movement of stocks but doesn't have any good reason for those predictions then it may very well move the price of those stocks in the ways that it predicts. But in doing so the money of the people who believe its predictions will be transferred to the people who don't until it stops having an effect.

Re: Using a Keras Long Short-Term Memory Model to Predict Stock Prices

#98

Post by jakeinspace »

sergioisidoro wrote: post #9

I preliminarily did the same thing as in this article a while back (with crypto data, since it's easily accessible), just as a way to learn keras. Somehow my network always learned to output a delay of the input no matter how hard I tried to shape it. I've searched through literature briefly and some examples. Some blog posts even claimed they had good time series prediction when I clearly saw they were having the sa…

Nobody's gonna do much better than lagging the market price using a model based purely on historical price data, markets aren't nearly that predictable.

Re: Using a Keras Long Short-Term Memory Model to Predict Stock Prices

#99

Post by T-A »

dointheatl wrote: post #24

11:15, restate my assumptions: 1. Mathematics is the language of nature. 2. Everything around us can be represented and understood through numbers. 3. If you graph these numbers, patterns emerge. Therefore: There are patterns everywhere in nature.

https://xkcd.com/1570/

Re: Using a Keras Long Short-Term Memory Model to Predict Stock Prices

#100

Post by zz34 »

goodside wrote: post #93

This is pure nonsense. This isn’t even the right way to begin thinking about this as a forecasting task — the target series should be log-normal returns, not raw asset price. The performance of this model is laughably bad, which is probably why he spends zero time evaluating its effectiveness. You could trivially get better forecasts than this by naively repeating the last-observed price . This isn’t ML. It’s cargo-c…

Welcome to ML in 2018.
Post reply on HN