...

[ceased-ebc]

.

[timestocome]

It is not complete. It is a work in progress. It's intended to serve as sample code for people working on their own predictive algorithms and as starter code I use for more complex projects.

Currently I'm looking into several other things, genetic algorithms, reinforcement code.. I had to step back and brush up on TensorFlow as I wasn't able to get Theano to build the next networks on my todo list.

If you're looking for a tool to predict the markets for you this isn't it. If I find the answer I'm keeping it secret, getting rich and buying an island somewhere.

[timestocome]

What are you using now for tools?

A friend went on a job interview for a local hedge fund. She came over with a tech test they hand out. It was on spreadsheets. I realized there might be a market here. (Spreadsheets!?!, are you kidding? ) So I started digging.

I found a few interesting things, the histograms of bears and bulls are very different. Bulls give a power curve, which is what you'd expect in finance or social data, bears look Gaussian. What I haven't found is a method ( other'n the obvious it looks like a Tulip Bulb Market ) indicator. I'm still looking for the new 'lipstick' indicator.

I had some short term prediction success with time series. Nothing earthshaking.

There's more of a herd behavior now, that always destabilizes and breaks things. The federal government being in a disarray isn't helping either.

Also on the to do list is to download a bunch of individual stock data and see what, if any, of the financial statement data predicts stock price movement. Maybe it's what we already know? Maybe there's something else in there? Clustering stuff is still pretty dodgy. I'm not sure how much I trust it?

I'm currently working on a robotics project, thinking a lot, and improving my TensorFlow, so it might be a few weeks before the Stock repository gets updated.

[timestocome]

Interesting ideas on Invacio, I love the ideas behind it, many of which I'm trying to look into on a smaller scale. Some of the website seems to be still under construction, but I wandered around the sections you have up and running. Banks are conservative, despite their reputation for being radical when ever the market crashes. That bot that bancrupted the hedge fund a few years back didn't help any.

I had a chance to look up the Ichimoku indicator. idk. I ran simulations on almost all the common ones ( that had formulas, not 'we know it when we see it' ) and some worked if you mapped them to a lucky starting day. Most don't work, or fail if you change the time window you test them on. I've tried a few ways to use machine learning to find new ones, I've not been pleased with the results. That's still a work in progress. I do think, that like Wolfram's simple rules that had complex results from simple rules, there are simple rules. I haven't found them yet. I have a stack of books on chaos and non-linear dynamics on my shelf, along with ones on genetic algorithms. There's never enough time.

I do think social movements and the market are very entangled and there are simple indicators from the herd behavior of the public that give information on future market movements. Then there's the ever present which drives which? Chicken first or egg first?

Much of what is happening in the US right now is fallout from the 2008/09 crash-bubble that that government totally mishandled. I think it's going to get much worse before it gets better.

Using large amounts of data can show us things we hadn't noticed before, there's been some huge break throughs, especially in medicine. There is also a danger that if you have too much data for the network you can find multiple paths that work and there is no clear way to know which one is the correct one and which path will lead you to ruin. ( see: derivatives, housing bubble ;p) Finding the sweet spot is the challenge. There'll be some more major screw ups because of this.

It's all fascinating.