Stock prediction using Weather, Twitter, News & Stock sites... | Hacker News Reader

Stock prediction using Weather, Twitter, News & Stock sites...

173.203.58.79

28 pointscaptaincrunch24 comments

While waiting for a few final things on my start-up I mentioned in an earlier posting, I wrote a simple stock prediction program in less than a day...

This program was created in 20 hours. It pulls in tweets, weather forecasts, stock market websites, current and breaking news, as well as looks as the trends of the stocks that it finds over time, to create the following collection of stocks. The stocks are rated based on all the factors previously mentioned. I am constantly tweaking each parameter, everyday, and the results seem to be getting better, and better. I will be releasing the source code over the next few days once I clean it up. If you have any suggestions on how I could make the predictions better, feel free to contact me at the email address below.

The entire project is written in PHP using MongoDb and memcache.

Here is the link: http://173.203.58.79/scatter/sc.processtext.php

Open articleView on HN

I'm just really amazed by the license a bit down on the page:

"BY VISITING OUR WEB SITE, YOU ACCEPT THESE TERMS, AS WELL AS OUR PRIVACY POLICY WITHOUT LIMITATION OR QUALIFICATION."

I mean, how could I read the damn notice without "accepting" it? Also, helpfully put BELOW the useful part, which you cannot "unsee" if you don't want to accept the license...

It was thrown up quickly, as I make things better, I will separate the annoying license from the goods. I just got really paranoid after making this type of site...

It's an amusing idea, but I'm not sure how interesting this is without a performance history or measures against similar automated predictions.

I'm going to start simulating buys/sells with history. Any reports that you feel would make it more amusing, let me know and I will add them ;)

Here it is: doing auto-trades with history.

http://173.203.58.79/scatter/sc.transactions.php

"Key: Symbol/(Perc. we estimate success)"

Not a very useful key. Maybe I'm missing something. At first, I thought maybe the colors were the prediction: green means up, red means down, but they just seem to reflect the percentage of "success".

What are the actual predictions? Is there a time frame for them? What do you mean by "success"?

I suppose I missed this when I whipped it all together.

The success is a percentage - the higher it is, the more ideal of a buy it is. The lower, the more chance it should be shorted.

The 16 hour predictions are more "short term advice" where the > 48 hours is more of a long term state.

It does not act on data older than 48 hours as things change to often.

Better not to give any suggestions to them as their license states:

>Please Do Not Submit Ideas, Suggestions, or Other Content to Us Unless Your Intent is to Give Us That Information Unconditionally

No offense, but in another thread that's at the top of HN right now ('Ask HN: Please explain short selling?), you're asking people to explain what short selling is. Based on that, I'm not sure I trust any predictions coming from this system, at least not on the sell side. This is a complicated problem you're trying to tackle, which requires a thorough understanding of the markets.

Measuring the psychology of the markets using web seems to be an interesting idea. This a better shot then most I have seen. Next step is to see if each stock's results are leading or lagging and over what time period.

What to the colors mean? There are various shades and also blinking colors -- it's maddening when you think you should focus on something, but then you have to figure out why you should focus on that....

The hugely long disclaimer at the bottom is rather distracting.

One can only assume, the creator has had some problems before with previous sites, or has had extensive legal dealings in offline world.

It sure is, I'd be rather worried having no disclaimer. How is this a bad thing?

Do your predictions work? Until you have some evidence either way, there's not much to say about this.

"Predicting" stuff is easy. The hard part is: do your predictions perform better than a coin flip?

If you actually spent 15 minutes to audit these picks you'll realize its a massive failure.

Prove it? My trades are often purchased and sold in 30 seconds.

Not often you see websites just going by IP. New trend?

no

what is the "percentage" rating? return predictions should be in return space.

someone should write a "how to write a trading algorithm" tutorial.

just read this recently found it ok: http://www.puppetmastertrading.com/blog/2010/04/08/pairs-por...

ernie chan's blog is also a decent starting point: epchan.blogspot.com

Oh, I know how to do this.

It'd just be nice to have a clear, concise guide to point to every time something like this comes up.

There's two parts: there is the strategy itself. And then there's everything that surrounds it: what a generic strategy looks like, how to test it, how to operate it, how to build risk models, etc. There's probably more work to the second part than the first.

nice.

Event arbitrage

Reply on news.ycombinator.com