All computing is predictive modelling - My Journal

A sparse multi-year journal towards profitable automated trading systems

In this journal I'll document my progress towards creating profitable automated trading systems. This is going to be a wild ride as there's a lot of work to do and I have a lot yet to learn, therefore this journal is likely to: last a few years, follow me across countries, survive through personality changes, and witness long periods of my complete withdrawal from it and the goal; I'll most definitely switch periodically to my other projects in search for quicker ways towards a grander goal which stays veiled for now. With this journal others will be able to get much needed new input to their brains, new ideas, while I am able to practice careful thinking and have a little fun laying out some of my work.

I currently have no experience trading. 4 years ago I committed to trading the currency markets and the first thing I've done is to learn programming, properly. You will see in this journal why I believe the brain is not adequate for intraday currency trading without some serious heavy-duty tools, and with them there's not much left for the brain to do so it's best to automate everything. The learning stage is still on: now in the math, logic, reason, and computing section. I need to learn these topics properly before I grind on creating automated trading systems (luckily, my other projects will also benefit from this knowledge). Expertise is not the same thing as experience, I can not just start creating automated trading systems or I get stuck in a local maxima just like many on here. It is fully necessary to think before you do, and achieve expertise before you see. Education is what enables you to set forward the necessary steps towards achieving your goals.

Errors and inaccuracies will not be corrected when caught unless they reduce the reader's experience. I'm applying new knowledge, it can be expected that I produce some errors. Also expect my writing style to be horrid for a long time as I do not have time to study writing. Now we begin. I'll start the journal with a thought exercise, the fruit fly:

I've never made friends with a fruit fly.

The fruit fly, it doesn't acknowledge me. It doesn't see me. I don't exist to it. But why? A fruit fly computes, doesn't it? A fruit fly sees, doesn't it? A fruit fly thinks, doesn't it? A fruit fly interacts, doesn't it? So why do I not exist to it?

A fruit fly landed on my leg: "AHA! There's my fruit fly, I knew you'd be my friend. Hello? Hmm I wonder why it's not acknowledging me."

I put food on my leg: "AHA! There's my fruit fly, back again. Maybe now you'll acknowledge me, I've given you food and everything!"

Impatient I became: "What is it with you fruit fly! You can interact with my leg. You can acknowledge my food. But me, the very character that's giving you the time of your life, the safest place you've ever seen, all the food you'd ever need, you can't even acknowledge!"

And that's the thing with fruit flies. They may have eyes like you, they may sense heat like you, but you as a person simply are a pattern too subtle, too hard to notice and handle. There's two reasons for that. One, the eyes of the fruit fly receive much less light than we are accustomed to. The fly is only going to notice patterns, structure, across way less light. Not many patterns exist there. Reason two, even if the fly did receive enough light, you are simply a constellation to them. You don't matter. You're hard to see. You're just some abstract pattern in a sea of light that has nothing to do with the constellation that is you. Or so they think!

"But I matter! I can crush this fly at any time I want." Exactly. The flies may currently be successful, many, and seem to go on with life just fine without noticing us. But if they did notice us, their accuracy in matching input such as light with optimal output such as movement would go through the roof. They could live like kings, following us humans around like the food and shelter producers we are, while being careful not to get too close. A subtle far out constellation and complete waste of time? Hah, not anymore! Oftentimes the most subtle patterns for you today may be your richest patterns tomorrow.

This is going to be a sparse journal. I am working on many things including my own programming language and libraries which strictly precedes the main part of this project and others.

Let's continue with the last post, I want to explain what I meant from a different angle now. This is certainly meant to be a journal but for now I am posting ideas. I have also yet to explain the title, but it will remain unexplained for now or I may just change it.

When you're doing something that's not supposed to be possible from many people's point of view, and you have little clue how to do it but you must somehow progress, it helps to find many different viewpoints to look at the problem from. It turns out one such viewpoint is the brain because brains excel at seeing structure in data.

The key is in the interface

In Australia male Buprestidae beetle have been observed trying to mate with beer bottles (Australian Journal of Entomology). The Paussus beetle is able to move throughout an ant colony at will and prey on the ants, never eliciting agression from the colony by imitating (through evolution) the sounds that ants make (Plos One). The sight of a displaced egg elicits a graylag goose's egg-retrieval behaviour, the graylag may behave the same way towards similarly shaped objects such as golf balls.

Why does this happen? Does a Buprestidae male beetle know the bottle isn't a female, but simply like it better as a mating companion because of its exaggerated features compared to the female bettle? Or does it, because of its small brain, rely on tricks that are more efficient than seeing a more accurate reality and thus actually see the bottle as a female beetle? Either way, there's not much of a difference: the beetle's decision making process is being told by its 'interface' to the world that the brown, shiny, and bumpy beer bottle is either similar to a female beetle, or that it is a female beetle. We, the humans see a more computationally expensive "reality", one where we see additional structure from the same light and tactile source.

Let's take a look at our interface.

Attached Image

See the square? Can you almost see the corners?

Gone now?

A badly drawn cube, or a badly drawn hexagon? (Idea from Donald D. Hoffman)

While you may think what you see is the reality, this isn't the case. You see a model of reality. So I ask you, should you see the image above as a cube or as a hexagon, or like in this case as both? When shouldn't you? Should we see 3d data from 2d sources, even when we have stereoscopic vision that is able to confirm the surface is indeed 2d? That of course depends on whether this data is appropriate, whether it works well or not to receive it. This is the kind of decision you have to make for your algorithms: should they see this, or should they see that? We know the underlying data, we know the results of the data (when backtesting), now you have to decide if it was appropriate for your algorithm, or your consciousness, to receive this input.

If you are unprofitable, is it really your algorithm's decision making process at fault for your losses, or could it be that your decision making algorithm's interface shows it a beetle for a bottle? Like a person that wonders why he isn't rich should ask himself if he's even given people a chance to make him rich, a person should get into the 'head' of his algorithms and wonder if he's even given them a chance to succeed, or has he shown them ants for beetles? How much are a few moving averages really telling you? How many ants and beetles are your moving averages merging together?

Is it really the case of making your algorithm go long when a moving average of exactly x bars goes over a moving average of exactly x+y bars? And is it really the case of tuning the fine details of this algorithm to become profitable? I say this algorithm is hardly getting any knowledge at all to work with, it's too near-sighted. It's focused on the exact details of a few calculations, instead of finding structure in the data.

It's important also to realize the biases that you must build into the system. Look up for example the hollow-face illusion, you see the depth wrong, and yet if you were to flick at it with your finger, you'd hit it exactly right. This is because different parts of the brain use a different interface to the world (I'll have to buy a hollow-face mask to try this some day). Similarly, you'll design algorithms that are forward thinking, eager, to see what really isn't there yet, which you'll tune to get a nice flavour of that behaviour (You as the creator of this system must constantly look at the input to your algorithm, and verify that the algorithm is reaching the conclusions you think it should). And then you'll build other algorithms that have the opposite effect.

Why would you bias an algorithm? Well you need to bias it towards seeing structure in the data. No algorithm, or brain, will do that on its own without being designed to do it, even though a brain can learn, it's evolution that does most of the work. Take for example current artificial neural networks, while theoretically they can model any function, they are honestly complete and utter shit because they are designed wrong. The human brain can learn from a single data sample while current artificial neural networks need tens? hundreds? or thousands? of examples of a watch, just to learn what a watch looks like. This, my friends, our brains don't need to do. I can look at a rare turtle just once and later recognize it from any angle, scale, and rotation (in a variable amount of time of course). This we will soon figure out how to do for ANNs/AI algorithms (see Vicarious for instance).

I hope I've given some food for thought, which is all this is meant to be. I'm currently working on my programming language which is turning out to be the most forward thinking programming language out there, by far. When you completely redesign the base, the fundamentals of what a programming language is and should be, a ton of incredibly useful features logically follow. More on that later.