Time series is an important part of financial analysis. Today, you
have more data, more data sources, and higher frequency of data. New
sources include new exchanges, social media and news sources. Today,
delivery frequency has been increased from dozens of messages every day
to hundreds of thousands of messages per second. Therefore, the results
will bring more and more analytical techniques. Most modern analytical
techniques are not different, and they all have a statistical basis, but
their applicability follows the available computational capabilities.
The increase in available computing power is faster than the increase in
the number of time series data, so it is now possible to analyze
large-scale time series in an unprecedented way. This neural network
predicts the future movement of the index and achieves a reasonably well
result.
Content:
It downloads the stock/ index data from an online information
provider, then forms a pandas DataFrame that contains open, high, low,
close and is compatible with the TensorFlow library and Keras. Finally, a
LSTM recurrent neural network will be implemented to train and predict.
It also creates a visualized result for the ease of presentation.
Optimized Hyperparameters arre also provided at the end.
Error
Please notice that I created this model with pandas 0.19.2 and I will update it to support pandas 0.20
Recently, pandas datareader is not available for downloading stock data, I am currently working on downloading data from Quandl.
How it works:
The efficient market hypothesis (EMH) states that price cannot be
predicted based on previous price and this model clearly violates the
EMH. It attempts to understand the market sentiment behind price trends
rather than analyzing a security's fundamental attributes. In order to
strengthen the market sentiment analysis, a sentiment analysis model or
event driven prediction model will be added. Hopefully, the result would
be slightly better than a random guess. The model is currently
overfitting and more updates will be provided.
