The length of the sequence and the number of nearest neighbors are free parameters. The sequence shall not be too short to get something which might have an interest. Too many neighbors make the prediction neutral (converges to the marginal ...

New historical Forex models

+1 Cauchy distribution has neither a mean nor a standard deviation nor any of the other moments defined. Yet a Cauchy i.i.d random walk is a random walk. The very short term (ticks) shows high antipersistance (but spread prevents making profit). ...

Trading Logic and Market Constants

By visual inspection it looked like an autoregressive model. I tried. I didn't use ACF or whatever. I trie with 2, 3 and 4. It was just a quick and dirty attempt. Some of the residuals are really big compared to the others. A linear model is ...

New historical Forex models

It is simply the index of the sequence of the swings. The Y-axis is the log ratio of swing height divided by previous swing height

New historical Forex models

I fit an AR(2) process on this ratio of zig-zag legs heights. At first it looks good (top) but the error when it is wrong is huge (bottom). I made it an visual quick and dirty indicator and I can't see any use of that... image image

New historical Forex models

To me a swing is a swing only when it is "confirmed". Like a candle only exists when it is closed. Can you get some stats about the duration (in bars) of the next swing (for you the live one)? If you can estimate both the length and the duration ...

New historical Forex models

I lack the time to give it a try. Thanks for posting your results. I don't understand your classification. To me after an upward swing only a downward swing can happen. I'm also confused by the prediction part. What are p and o? What is length(p)?

New historical Forex models

I take all the sequences of N (=40) bars and search for the K (=20) nearest past sequences to the current one (K nearest neighbors algo). The metric I use is simply the Euclidian distance. The search is made through a VP-tree that is built once. I ...

New historical Forex models

Do you mind drawing a little sketch? I sounds like you need the next swing to draw the current triangle.

New historical Forex models

Here is your previous screenshot of E/J with the 20 nearest price action (of E/J M30, L2 distance) and their extension in the future. The red and blue strong lines are respectively the mean and the median of the extensions in the future. As you can ...

New historical Forex models

I keep searching how to use HMM for market conditions estimation. The only resources I'm able to find are only about the volatility. For example this one: url When you take the picture of his results and add a few lines onto it you clearly see ...

Machine Learning with algoTraderJo

I discarded Ornstein–Uhlenbeck process fitting for two reasons. The first is that no mean reversion was found in high frequency data by academics, quant oriented people in this forum and myself. Second I think the market choppiness periods change ...

Choppy market index: any good ideas?

I focused on the estimation of the variance because this is the usual definition of volatility (or standard deviation to get the same unit as the price). But I realize that's not enough. Here are two series I hand made. On purpose the blue one looks ...

Choppy market index: any good ideas?

I couldn't compare the distributions when the market is ranging/trending/choppy because I really lack a good estimator for these states. What I can tell is that overall the duration (in sec) of a range bar is distributed like a Possion and the ...

Choppy market index: any good ideas?

This is exactly the kind of classification I'd love to get. I would also add a fifth "crappy jerk price action" state where no zig-zag is apparent.

Choppy market index: any good ideas?

Your trading system can be a quick scalper or a long term strategy. You may swing or prefer breakout. You can pyramid or average down. You can take partial profits or "nhedge" your positions. You can trade based on TA or FA. All these different ...

Automated trading algo generation

The vast majority of the automated trading systems I saw are all about predicting where the price will go next. It can be the next bar or a few bar horizon. Let's say that over the next period the price is predicted to go up. Is it really a good ...

Automated trading algo generation

Thanks for posting PapaMack. The immediate counter-example I think about the HHHL idea is a powerful directional move. It isn't choppy at all and has no alternating highs and lows. I've tried various variants of this idea of comparing the market ...

Choppy market index: any good ideas?

@merquise I think it is important to understand the difference between correlation and cointegration. Whereas everybody has a good grasp on what correlation is, cointegration is not a word commonly used. They look like they are the same thing. If ...

Market Neutral Corner

Again a high correlation coefficient is not sufficient. Neither is it even necessary! This picture shows two counter examples. The two series on top have a correlation coefficient of 100%. They will never meet again. The two series at the bottom are ...

Market Neutral Corner