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

141–150 of 178 posts

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

#141

Post by goodside »

ak39 wrote: post #129
jpeterson wrote: post #114

Earlier quoted context omitted.

Investment markets operate on relative gain, not absolute gain. E.g.: if you invest in a stock and it gains $5, this would be a great return for a $1 stock but a poor one for a $1000 stock, so the absolute gain doesn't mean anything on its own. A 5% return always means that you've gained 5% on your investment.

Would it be the same (valid) with percentage returns?

Less so. Log-normal returns are better because they have the property that a summation of log-normal returns over contiguous intervals is equal to the log-normal returns of the combined interval. In other words: Losing 5% and then gaining 5% doesn’t put you back at exactly 100%, and log-normal fixes that.

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

#142

Post by joe_the_user »

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.

If enough people believe a given model and use it, it would actually tend to make the model over and under-predict. If a lot of people think the market will go up, it's race to buy before the others and eventually a race to sell before the others when either expectation shift or all the buyers run out.

Which is say, belief in a direction can make the direction happen but belief in a particularly shaped curve won't make the market resemble that curve.

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

#144

Post by java-man »

bobosha wrote: post #60
java-man wrote: post #57

It's tough to make predictions, especially about the future. -- Yogi Berra

Have heard this quote attributed to einstein.

Or Bohr:

[0] https://en.wikiquote.org/wiki/Niels_Bohr

[1] https://quoteinvestigator.com/2013/10/20/no-predict/

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

#145

Post by ipsa »

vl wrote: post #113
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…

Even more so since test data can't be from the same time range: i.e. for time series you need to split train/test by date, not randomly, otherwise your model just memorizes the series.

The article splits in time, not randomly.

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

#146

Post by auntienomen »

woah wrote: post #96
lifeformed wrote: post #77

Earlier quoted context omitted.

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

The fund contains only employee money. That's a very clever ponzi scheme if it is one.

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

#147

Post by poulsbohemian »

hhmc wrote: post #47
steve19 wrote: post #39

Earlier quoted context omitted.

How does a lifetime non-compete work?

I'm shocked that would be even close to enforcable.

I don't know anything about the finance industry, but it seems to me that there are a number of industries where the non-compete term is irrelevant, as you can still become persona non grata. For example Companies B, C, and D know that if they hire someone from Company A, they will get sued. The merits are irrelevant if Company A can outspend them relative to the value of the prospective employee.

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

#148

Post by modeless »

Loughla wrote: post #131
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…

As someone outside the tech sphere on either coast, that's all ML seems to be. What I've seen from companies marketing to Higher Education is - we have a lot of data, you set arbitrary flags to the data that you believe indicate 'x' (or even better, they have pre-built data expectations) and you will get 'y' outcome. And none of it is actually based on anything real. It's all anecdotal applied to extreme amounts of a…

> that's all ML seems to be.

You have to be very selective about what you consider "ML" to come to that conclusion. There has been a constant parade of incredible, mind-blowing results out of ML over the past decade, advancing the state of the art by leaps and bounds both in research and in real applications.

Do you not remember how terrible speech recognition and speech synthesis were just a few short years ago? Did you not see DeepMind finally crack Go? Check out BigGAN [1]. Try search in Google Photos. See translation getting better every year.

Yes, there are quacks and charlatans and people who are just plain wrong. But ML is real, and it solves real problems that people failed to solve any other way despite decades of concerted effort.

[1] https://medium.com/syncedreview/biggan-a-new-state-of-the-ar...

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

#149

Post by ChristianGeek »

lvs wrote: post #139
Loughla wrote: post #131

Earlier quoted context omitted.

As someone outside the tech sphere on either coast, that's all ML seems to be. What I've seen from companies marketing to Higher Education is - we have a lot of data, you set arbitrary flags to the data that you believe indicate 'x' (or even better, they have pre-built data expectations) and you will get 'y' outcome. And none of it is actually based on anything real. It's all anecdotal applied to extreme amounts of a…

It's so much worse to see ML applied to problems that could surely be studied and understood mechanistically, but people are sold on using ML instead in deference to buzzwords alone. The result is that they may get some model of a phenomenon, but they'll never learn a goddamn thing about why it works. What good is that?

It depends on what your goal is; sometimes it’s the destination, sometimes it’s the journey.

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

#150

Post by auntienomen »

T-A wrote: post #95
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/2017/05/0…

Magerman was an early employee. Not sure his agreement would be typical.
Post reply on HN