B is correct. The mistake is to say that "if I copy-trade that strategy with a risk multiplier based on account equity of 4x, this will be the expected outcome ...". The gain and DD aren't simply multiplied by four. With any system, at some point ...

How risk multiplier really affects monthly gains and drawdowns

It comes from your assumption which is wrong. If you win x% and then lose x% (or vice-versa) the result is not BE. Even with 1% you get a negative drift. Say you start with $10000 win 1% then lose 1% your balance is now 9999. It looks like almost ...

How risk multiplier really affects monthly gains and drawdowns

When you trade on margin that's your broker who (technically speaking) lends you the CHF not the SNB. Brokers decide the rate they want to offer. Strangely they decided to not pay you to borrow their money ^^

Are Negative Interest Rates Useful for Carry Trade?

Not only it can but when it does it's time to have a look on the cross without this currency. If N/U goes down and N/J goes up it means that NZD is weaker than USD and JPY is weaker than NZD. Therefore JPY is much weaker than USD => U/J trends ...

Currency Pairs - Same Currency, Different Direction?

Let's imagine that you expect the market to turn in a short time with high probability. It can be because of your new SuperUltimateOscillator_v34.ex6(r)(tm), a very strong S/R on the weekly chart, an insider or whatever. You don't exactly know where ...

Expectancy Management

From your assumption of the term random the short answer is: being lucky. But random doesn't imply i.i.d.

Profiting from an unpredictable market?

I understand that. The problem is that you select only a set of successful random trading sequences. This is called survivorship bias.

Proof of Profitability

You do so implicitely by using the formula. It assumes no edge at all. The average profit you make is the sum of the money you make when you win and the money you lose when you lose, which is negative. You win with probability P and lose with ...

Proof of Profitability

The flaw #1 is in the first post. When you write P(TP)=SL/(TP+SL) --I fixed the typo-- you're assuming a random walk with no edge. Therefore, per definition, you can't have any strategy even slighly profitable. From there, there is no point in ...

Proof of Profitability

Hello, I'll try my best to clarify These are the rules you set on top of the prediction. In the examples so far they are not infered from the data. You decide upfront. Up to you. You can average the K nearest neighbors predictions. If the result is ...

Machine Learning with algoTraderJo

In your thread you say "some incredibly promising open-source software" and "using their proprietary algorithms". Either it's open or it's not. What is the license? If it is open source can you point us to their work to let us asses how incredibly ...

Machine Learning with algoTraderJo

Random is not the opposite of predictable.

Random vs Predictable

Unfortunately it is far from that easy. I could explain why it is wrong but I think that if you're reading this thread you'd like to learn about machine learning for trading. That's why I think it is better if instead you explain why it isn't the ...

Machine Learning with algoTraderJo

In this downtrend 42 out of 210 bars have not been fully revisited (the whole range not only the open price). That's only 20%. image

Leftovers at the dinner table, re-testing opening levels

Did you have a look at Sugeno fuzzy logic models? To me it looks like this is what you're looking for. In fuzzy logic the boolean operators AND, OR and NOT become functions. For example x OR y may becomes x+y-x*y. There are several formulas. Note x ...

How to add weighted conditions in trading formula

Not only we have very few samples but the data is very noisy. Let's see with 50 samples and 20% chance a sample is mislabelled: image image

Machine Learning with algoTraderJo

For trading purpose we have very few samples compared to the universe of the possibilities. Here is ELM with 30 neurons vs Gaussian Processes. There are now only 50 samples and no mis-labels. I much prefer what the GP finds. image image

Machine Learning with algoTraderJo

50 and 100 neurons. With too many neurons it starts learning the dataset rather than explaining it. It takes care of the exceptions too much. From what I understood of the paper there is no selection of the important coefficients. On the contrary ...

Machine Learning with algoTraderJo

The same with 500 training samples image image

Machine Learning with algoTraderJo

No I don't use this. Before your post I didn't know it existed ;-) I downloaded the java version and tried with a toy example: using 100 samples I make it learn if a 2D point is above or below a function. For each sample in the training set there 5% ...

Machine Learning with algoTraderJo