News articles and trades ideas should be posted with your analysis or a question. Post an article, or link to multiple articles with your analysis in it. Also, concerning your analysis, see The following should go without saying There is a thin line between promoting a product and service and SPAM. Before you can post a promotion or service you need to get approval from the mods. Links to external services, especially if they are selling a product such as signals services or binary options will neural excised with haste. Want to post a trade? What inputs would you choose for an intraday neural network? Im not sure how many algo traders network this sub but in any case all ideas are welcome. I am in the process of developing a neural net for the EURUSD and would be interested in getting some idea for the inputs. I am somewhat new to algo trading network have been spending last couple of months teaching myself machine learning, R programming and now recently focusing on the theory of neural networks. Next step is to start playing around with different versions of neural nets different number forex nodes, different inputs, different architecture etc. The complexity is still high and I have much to learn. However, I would be interested in hearing from experienced traders what inputs they would choose for a neural network? Identifying inputs is probably one of the most difficult things in developing a neural net and to have a variety of ideas to start from would be a great help. For those unfamiliar with neural networks they can briefly be defined as a type forex black box strategy which takes x-number of inputs and turns it into y-number of outputs through a learning algorithm. In terms of selecting inputs, you need to spend some time "chart time", meaning time analyzing the indicators and their relationship to the price movement. Don't necessarily look at just the indicator value but how you are actually reading them. Are you looking at the slope of the RSI? The distance between two moving averages? Whether the MACD Line is positive and above the signal line? You want the values you feed into the NN to contain as much information as possible and calculating features off the indicators is a network way to do that. A couple indicators that I have neural success with are: CCI, MACD the histogram forex particularADX trend strengthmoving average crosses especially EMAsAROON, and DeMark Trendlines a little more difficult to quantify programmatically. The testing process is incredibly important. Just splitting your data into a training, test, and validation set is not enough. You also need to decide on the criteria you're optimizing for. Is it profitability, accuracy, some risk-adjusted return metric, or stability? Remember that the entry signal which is what you're finding here is just one small part of the equation. The list goes on. You also need some foresight into how you are going to actual use the NN to trade live. Finally, define what success means to you. Set concrete goals for your strategy not just the NN and try to achieve them. Even shooting for 20 pips a week can add up over the long term. You need to know where you're going in network to get there. The list can sound daunting but take it one step at a time. If nothing else, this will be an incredibly valuable learning experience with some serious upside. Happy to help out more if you have any other questions or would like some input on how you're approaching it. I also friended you and will probably take you up on your offer and be in touch if i get stuck at some point. Meaning it is more difficult to find strong, reliable signals. That being said, there is a benefit to having more data for these types of applications finding clean, reliable data is another step you need to take btw and you should build a strategy that fits your profile. I would recommend starting with 1h data 4h is even neural but if that is too long for you, I wouldn't go less than 30min to start. Another forex to consider is the more you trade, the more transaction costs are network to come into play. You should be incorporating transaction costs in your returns from the forex seeing the effect of transaction costs might make you reconsider your short timeframe idea. It is difficult to generalize that more trades will lead to quicker recoveries from drawdowns and my intuition is that more transaction costs will more than offset any advantage there. My only suggestion is to start simple and work your way up from there. In my experience, the choice of inputs and how you are utilizing the output are far more important than the structure of the NN unless you are really getting into the deep learning field or are interested in unsupervised learning. Get the process working from start to finish with a basic back propagation algorithm, using a logistic sigmoid activation function and one forex layer, then build from there. I would start with a single output: I haven't tried any of them but it looks like there are some R DLLs for MetaTrader 4 that would let you use your models in the platform. For the validation set, you run the risk of building many different neural nets and only selecting the one that performs the best on the validation set. You are effectively fitting the model to that particular dataset meaning it won't generalize well. There many different techniques, like cross-validation techniques or using a sliding window approach, that you should look into. Good luck with getting this up and running! Take it one step at a time, get it working as simply as possible, and only add complexity when you are seeing a high ROI. It should be as simple as possible but no more! The guy is not interested neural chart time Learn to trade manually first. Become proficient at a simple strategy first. Become profitable and only then neural about using such tools to HELP you trade not trade for you. Trading is easy once you know how. Do not complicate things by chasing some magical new way that will make you trillions. Algo trading is advantageous because it is more repeatable and measurable. Its very hard to backtest a manual trading strategy because you cant backtest emotions and peoples reactions to specific situations. To the OP, Identifying inputs as you mentioned is one of the most difficult parts of creating a forex strategy. As such, no one is going to give you a good answer to this question. That is like asking someone to spill their secrets to you. I've been running an algo strategy for 2 years. Leverage varies from 1x-7x usually maxing out at 9x. I have NEVER traded manually before this. Just because it didn't work for you doesn't mean it won't work for others. I trade in a fully automated way and do all right. Sure looks like it. Maybe I should be more explicit: I trade in an automated way that is in no way about using automated tools to help me trade manually. I literally don't neural the first thing about trading manually in a way that's relevant to these strategies. Companies dropping millions to phds for algos and you are asking tips on reddit. Not saying you are doing wrong but this situation looks weird. I hope you find your success. I am a bot, and this action was performed automatically. Please contact the moderators of this subreddit if you have any questions or concerns. Use of this site constitutes acceptance of our User Agreement and Privacy Policy. By signing up, you agree to our Terms and that you have read our Network Policy and Content Policy. Log in or sign up in seconds. LIST OF RESOURCES CHATROOM SO WHEN WILL YOU BE RICH? No empty news articles or analysis News articles and trades ideas should be posted with your analysis or a question. Also, concerning your analysis, see 2: No Spam, No Non-Approved Promotions There is a thin line between promoting a product and service and SPAM. Keep It Professional No insults or attacks of any kind. Abusive posters will be network. Posting Trades Want to post a trade? Recent USD trade I took: This is an archived post. You won't be able to vote or comment. Im an IT guy, I code and Ive written Backpropagation engines from scratch. I wasted years trying to find an automated way to trade. I hate comments like this. The problem with "black box" strategies is that they are notoriously hard to debug. Please ensure you are providing quality content. All neural will be reviewed by the moderators and appropriate action will be taken. It has been removed and will be reviewed by the moderators. Remember, Spam will not be tolerated. Posts are automatically archived after 6 months.
.jpg "forex neural network")
Both views must be taken into account, but neither is adequate.
Japan changed its system somewhat in July 2012, but the card system is essentially.