I did a job on vWorker which was implementing a trading algorithm. The guy had a system, was using it to trade and had a paid service for people to buy it from him and had a paid subscription video channel of videos of him trading the previous day.

When I used what he thought were signals he was following (exponential smoothing, direction finding and spread detection iirc) it generated plenty of "false" trades.

It ended up he refused to pay because "obviously" my code was applying the rules wrong, not his rules didn't work.

I eventually got 50% of the fee and learned quite a lot in the process of why trading algos will struggle to work when faced with real random walks.

>It's challenging enough to translate simple trading systems into machine code

I can attest to this from my experience with a bot I'm trying to build to handle my personal finances. When I sat down to try and encode the seemingly simple decisions I make like whether to pay a bill now or wait until the next check, things got really complicated really quickly.

I'm struggling a lot to get access to my bank's "bill pay" interface. Any recommendations for banks that support building out automation like this?

There are very small things I'd like to implement (budgeting, auto-categorization of expenditures) and some type of bill pay as well.

<strike>I can't really help you on bank API's</strike>. My bot uses webdriver.

Actually I can help with that: https://openbankproject.com/for-developers/

Huh.. Looks like they don't support my bank, but really love the project!

I'll have to check out webdriver, thanks!

Not to mention the fact that there is incentive for the people you are trying to use to try to mess you up. There are so many little mistakes you could make that would result in significant errors when added up that it would make the end result worthless. And it's in their interest to do that... the employees.

not necessarily. some (maybe most?) black box quants own their algorithm even if they leave the Hedge Fund. And of course they make profits while the algos are used. So the architects of the system are incentivized for it to work well.

> some (maybe most?) black box quants own their algorithm even if they leave the Hedge Fund.

I've never observed this be the case at all.

It's the default case in the USA, not sure about other countries. The contract has to specifically sign over ownership.

this was the case at 2 funds I've worked at. So I assumed it was the norm

Vindicis, unless you're not working, you probably already work at a place like this.

All Dalio is doing is tying the pieces together and stating the obvious.

What if I am working, but not at a place like that? I know what you mean, but that's a very large and inaccurate assumption you're making there.

Stating the obvious, and implementing it, are two very different things.

If it were so easy to model and implement human behaviour and thought process, we'd already have strong AI. As we don't, perhaps it isn't quite so simple, regardless of how obvious it might seem?

The point I was trying to make is that when you're trying to model that sort of thing, it very quickly goes from "Seems easy enough." to "Great time to start a new career!"

The complexity of these systems ramps up exponentially.