Neuro-forecasting of financial series (based on one article) - page 3
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
This work is not on the level of our PhD. A bit cooler than a diploma, but far from a PhD.
MS or MA - it is cooler than our "Specialist" (training lasts 4 + 2 years, vs. 5 years for a specialist), but falls short of a PhD (our PhD). But which is closer, the Specialist or the PhD, I don't know.
If the participants do not mind, I will try to fill it with articles (cases) of successful implementation of neuro-forecasting in financial markets, including forex. I think it at least will not make the article less informative, and something may hit the nerve of fans of neuro (how it sounds).
So, Noisy Time Series Prediction using Recurrent Neural Networks and Grammatical Inference: http://clgiles.ist.psu.edu/papers/MLJ-2001-finance-time-series.pdf
Working with daily forex quotes. At first BP is converted into a pattern, then pattern-samples are made (i.e. dipping into the lag space to the depth of 1, 2), a SCN is trained on this array in order to break the array into a finite set of alphabetical characters (belonging to SCN neurons, number from 2 to 7). A recurrent NC is then used to predict the next symbol.
Results: by increasing the confidence interval boundary to Percentile 0.95 (i.e. including in the analysis only 5% of the strongest NC outputs), the prediction accuracy of the daily closing price increments is around 60%. At the same time the volume of test sample is reduced to 225, which by the way is not bad either.
In my opinion, it's a very high quality article, the main pitfalls are outlined, the conclusions are not striking, but only slightly surprising. And the ideas are quite progressive (meaningful, ideologically nourished, nurtured...)
On half of the trend we teach, on the other half we predict. In this article there is a long trend section in all pictures. It was not the price that was predicted, but the direction (either yes or no). At the end of the trend we can see price spikes and no forecast on those spikes. Two MAs will show the same results on the data section shown in the article.
100% trained as yesterday and waves have nothing to do with it. Believe me, a seasoned tester of all sorts of NS. And the concept of philosophy of what NS is....
100% trained just like yesterday and the waves had nothing to do with it. Believe me, a seasoned tester of all sorts of NS. And the concept of philosophy of what NS is....
I highly doubt that the trend has anything to do with it. Even in the case of a trend, the price movement sign alternates randomly, and the article predicts precisely the sign for a step forward.
On half of the trend we teach, on the other half we predict. In this article there is a long trend section in all pictures. It was not the price that was predicted, but the direction (either yes or no). At the end of the trend we can see price spikes and no forecast on those spikes. On the data section shown in the article two MAs will show results just as good.
Do you think that on a trend, two wands will give the same result? Have you tried predicting the direction of price one day ahead? The result will be 50/50 (maximum, 40/60) on a sample of at least 150 days, whether it is trending or flat. I understand if we hold the deal from crossover to crossover, in a trend. We are talking about something else in the article.
PS: If I am wrong, throw a stone at me. Not the first day married myself.
I highly doubt the trend has anything to do with it. Even with a trend, the alternation of price movement signs is random, and the article predicts exactly the sign one step ahead.
I don't even care what anyone predicts, I admit I haven't even read the article. But what they predict there (the sign, the direction, if I understand it) with a 94% probability is IMPOSSIBLE!!!!!. Believe me. It's not usually higher than 64% for a prediction, and that's!!!!! That's a hell of a score. And you're saying it predicts candle colour with a 94% probability!!!!! The ns can't do that. I admit honestly I have not programmed the net, but as a user of it, I have. NS is not a magic key, unfortunately. In order for the network to work, you have to send at least one signal to the network input which works even without the network, only badly. Let's say it doesn't dial. It holds, but it does not dial. So using NS together with another indicator is POSSIBLE!!!!!! Not 100% sure that NS will help make the signals better. BUT..... NS will give you several ways of training, but no one knows which of them will work. You can only hope for luck that you will choose the right one.
In any case, I have never met an NS which gives 100% confidence in the future. At least 80%. All the more so when it comes to predicting a sign, with a 94% probability it's just not possible. Not even one step ahead....
Do you think that on a trend, two wands will give the same result? Have you tried predicting the direction of price movement a day ahead with a dab? The result will be 50/50 (maximum, 40/60) on a sample of at least 150 days, whether it is trending or flat. I understand if we hold the deal from crossover to crossover, in a trend. We are talking about something else in the article.
PS: If I'm wrong, throw a stone at me. It's not the first day of marriage myself.
it's just that there is some cyclicality (momentum) ... so if this dynamic gets into a model (ns, regression, etc ... basically it makes no difference) and then continues for a while ... then you're good ... if it changes, then you're out ... reasons for changing dynamics 2...market and DT filters...
In any case, I have never met an NS which gives 100% confidence in the future. At least 80%. All the more so when it comes to predicting a sign, with a 94% probability it's just not possible. Not even one step ahead....
I haven't come across either, although as a user I've been intimate with neural networks for a number of years.
That's not the point. Just saying "it's a trend" is wrong in this context. And predicting a candle colour one step ahead with almost 100% probability is, in my opinion, almost fantastic. But where is the constructive criticism of the article? Or is there not enough data for criticism? Or maybe the author of the article just fooled everyone, even his esteemed scientist.
You should at least read the article, by the way. There, in addition to fantastic results, interesting techniques are used. I myself have not tried many of them.
100% trained just like yesterday and the waves had nothing to do with it. Believe me, a seasoned tester of all sorts of NS. And the concept of philosophy of what NS is....
I don't understand your position, do you think this work shows reliable results? Do you think it is possible?
I don't even care what anyone predicts, I admit I haven't even read the article. But what they predict there (sign, direction, if I understand it) with a 94% probability is IMPOSSIBLE!!!!! Believe me. It's not usually higher than 64% for a prediction, and that's!!!!! That's a hell of a score. And you're saying it predicts candle colour with a 94% probability!!!!! The ns can't do that. I admit honestly, the network is not a programmer, but as a user of it, it is. NS is not a magic key, unfortunately. In order for the network to work, you have to send at least one signal to the network input which works even without the network, only badly. Let's say it doesn't dial. It holds, but it does not dial. So using NS together with another indicator is POSSIBLE!!!!!! I am not 100% sure that NS will help make the signals better. BUT..... NS will give you several ways of training, but no one knows which of them will work. You can only hope for luck that you choose the right one.
In any case, I have never met an NS which gives 100% confidence in the future. At least 80%. All the more so when it comes to predicting a sign, with a 94% probability it's just not possible. Not even one step ahead....
I totally agree ... if you don't have good predictors you won't get any good results )))) With 1 VP it's hard to do anything good... and if you have predictors, you won't need all the mess with grids and other algorithms ...)