Stock charts are the graphical representation of a series of prices of any specific stock. Charts showcase the movement of the stock price over the limited period of time.

Each chart tells the story. The expert and experienced traders can successfully leverage the stock charting data to make intelligent technical analysis and trade better.

Advanced technologies like deep learning and machine learning can further be advanced the basic stock charting solutions by powering them with predictive analytics.

Discovering the depths of stock markets is made simple with deep-learning based stock-charting and risk prediction solutions.

We were approached by a popular Australia-based stockbroking firm to build an AI-based stock charting software solution. Their firm relied on an outdated product that was not capable of catch the dynamic market moments through charting analysis. Without the support of insightful data, their technical analysis was not up to the mark which in turn incurred financial losses and a major drawback in market reputation. The client firm wanted to scale their trading efforts through advanced technologies that can provider next-gen and qualitative technical analysis.

Our first goal was to evaluate how deep learning can be applied to identify common chart patterns from historical stock data.

We started collecting the financial data and identify the crashers.

The predictive analytics system architecture was designed to generate 30, 60 and 90 day forward estimates to examine the best time period window to predict near term crashes.

Our next step was to calculate price drawdowns, that notify the persistent decrease in price over consecutive days from the last maximum to the next minimum price.

To make the stock price prediction on any certain day in the future, the daily price changes from each day before the chosen day would be used as a feature. Our goal was to capture the essence of past price movements accurately at any point in time. To achieve this, we defined 8 different time windows to measure mean price changes. OSP leveraged a Convolutional Neural Network (CNN), a feedforward network which reduces the input’s size by using convolutions. We performed logistic regression and analyze the regression coefficients to evaluate feature selection.

Our stock charting solution evaluates the change in the ratio of the probability of crash vs. no crash.

We elected the S&P 500 data set for testing for being the largest data set with the largest amount of crashes.

We leveraged the Log Periodic Power Law (LPPL) approach to calibrate the financial bubbles especially to predict the time of bubble ends.

A crash or change of regime is by targeting observed price time series with LPPL approach increase the prediction accuracy.

For training, we would perform 6-fold cross-validation along with the use of the LPPL model. This means running each model six times and using five data sets for training and the remaining one for validation. To train our ML model, we performed 6-fold cross-validation along with the use of the LPPL model. The logical regression models have the potential to identify the optimal coefficients for a function by minimizing the difference of the predicted and the actual target variable.

Stock risk scoring is made possible with weighted harmonic mean of precision and recall called F-Beta Score.

The beta parameter determines how precision and recall are weighted, where a beta larger than one prioritizes recall and a beta smaller than one prioritizes precision.

The difference between the predicted and the actual target variable was made possible with logical regression models.

The problems with time series data are solved through loops that connect cells. At each time-step, the input is based on the output from the previous time step.

Our application of Long Short-Term Memory networks (LSTMs) helped to overcome the learning long term dependencies in RNN. The solution helps to predict the future performance of any particular stock with pre-defined time series sets like 7 days, 15 days, 30 days and 45 days. The smart integration with real-time stock trading and historical trading data helps to monitor, identify and analyze the stock performance with deep learning technology. The future performance prediction of stock helped our client to trade with success.