zbanga/fxcm_currency

Using Machine Learning for live currency trading

FXCM_currency

Using a simple machine learning model with a Logistic Regression model to predict and trade currencies. Currently it is on fixed/static time prediction (next bar), improvements will be used for varying forward time window prediction(Instead of using 5 min bars to predict the next 5 mins, aim to predict longer ahead)

Another interesting idea is trying to predict over an amount (basically transaction costs). Will aim to bin the data after minusing the transaction costs.

Example: https://github.com/SolbiatiAlessandro/RNN-stocks-prediction (dynamic prediction).

Also will aim to look at or account for variable change in prediction, say you in are currently in a trade but new data (or news) comes in suggesting that the current prediction is no longer accurate (try to minimise position or close out the current position).

Live trading is being implemented and working rudementary (trade at your own risk)

Documentation for FXCM API: http://fxcmpy.tpq.io/

Adpated from and credits go to : http://hilpisch.com/fxcm_ai.html

Download or install jupyter notebook to view files

Add your token from FXCM to the fxcm.cfg to start trading (Create a demo account and login to https://tradingstation.fxcm.com/ and find the token (should be under account)

Unique implimentation of this is binning the data according to voltility

Results:

Out-of-sample:

Out-of-sample with transaction costs:

Assuming FXCM active traders costs

In-sample:

fxcm_currency's Issues

Backtester doesn't account for trading costs

Need a more realistic way of accounting for transaction cost and slippage

Make Pickle

Save model in pickle form and keep track of it

Currently trades open but no way of tracking when they close

Trades open long/short but no way of tracking individual trade performance or closing them.
If a no-hedging account option is selected then its impossible to close currently.

execution error

System: Ubuntu 16.4
python: 3
running in jupyter Notebook
File: TripleBarrier.ipynb
`NameError Traceback (most recent call last)
in
----> 1 data_in[cols].iplot()

NameError: name 'cols' is not defined`

File: Automated Algo.ipynb
`eyError Traceback (most recent call last)
in
----> 1 data.iplot()

~/anaconda3/lib/python3.7/site-packages/cufflinks/plotlytools.py in _iplot(self, kind, data, layout, filename, sharing, title, xTitle, yTitle, zTitle, theme, colors, colorscale, fill, width, dash, mode, interpolation, symbol, size, barmode, sortbars, bargap, bargroupgap, bins, histnorm, histfunc, orientation, boxpoints, annotations, keys, bestfit, bestfit_colors, mean, mean_colors, categories, x, y, z, text, gridcolor, zerolinecolor, margin, labels, values, secondary_y, secondary_y_title, subplots, shape, error_x, error_y, error_type, locations, lon, lat, asFrame, asDates, asFigure, asImage, dimensions, asPlot, asUrl, online, **kwargs)
1156 sharing='private'
1157 if sharing is None:
-> 1158 sharing=auth.get_config_file()['sharing']
1159
1160 if not filename:

KeyError: 'sharing'`

Running the code

Can you walk through running the code? Is it just copying this code into a python script? If that is the case, my results after commission are much lower than the images you show.