Search results

Well, I can choose to use whatever structure in terms of number of layers and hidden neurons. The question is do you think there is SOME structure that will allow the NN to pick-up this relationship? And, what do you suspect it might look like?

5 days I am using at the moment. And currently I have 3000 rows.

Great! We are on the same page. My testing confirms that is does NOT matter. Here is the next "thought experiment". I am trying to predict the future price change FPC. If I supply the FPC as an input, then the NN uses this information and makes pretty-much perfect predictions. No surprise...

That's exactly my line of thinking! I believe you should get the same answer when input is 8 - in each case.

It was your second example: 8,6 9,8 13,9 11,13 10,11 And I can choose to use between 100 and 3000 training samples

Yes, that sort of thing

1 day forward change

I train on the first part of each dataset independently, but with the same inputs (except reversed data of course)

I am just reversing the bar order - no other changes. And Stock A is in a strong rally through the period, and therefore Stock AR is in a strong decline. I would expect the same performance because the only thing different is that the sign of the changes I feed to the neural net have flipped.

This is a NN, not a MA breakout type rule.

Here's an interesting thing. I have created a simple neural net with 1 input (20 day price change) on a stock A that go UP over time. I then created artificial data AR by sorting the prices in reverse time order i.e the stock now looks like it is gone down. My intuition tells me that the...

What accuracy are you targeting? 70% actually sound very high. AlphaGo only get 57% accuracy on matching a human's decisions. And many trend-following CTAs have win-rates in the low 50%.

I want to trade my simple neural net across 20 stocks. In simulation I could Train "walking-forward" based on a single stock - train for x months, paper-trade for y months and keep rolling this window forward. Or I could train across all stocks in each window. Do people think it is best to train...

I have introduced early stopping into my system. I do this by measuring the slope of the recent values of MSE over the past x iterations. However, this gets triggered very rarely - I had assumed it would trigger often and I would save time.

Of particular interest would be comment from somebody actually running a NN for live trading. Across a portfolio of currency pairs or stocks.

Perfect

Getting back on topic. Has anyone managed to build a simple NN that gives a reasonable trading performance across a set of instruments?

For comfort, I think it should work on out-of-sample data of 2 years. AND it should work across a set of instruments (say 5+) that have been selected PRIOR to testing the models on them. Selection bias in the instruments that are chosen for modelling is a very easy/common mistake.

I am sure I am misunderstanding you when you say 100% accuracy. Can you clarify what you meant please.

OK thank you. One issue I grapple with is that it is relatively "easy" to create a model (simple heuristic - not NN) for a particular Stock. When I try to apply the same model to a range of Stocks it often does poorly. I usually assume that the Stock it worked for was just fitted. Do you have...