Machine Learning for Intraday Stock Price Prediction 1: Linear Models

This is the first of a series of posts on the task of applying machine learning for intraday stock price/return prediction. Price prediction is extremely crucial to most trading firms. People have been using various prediction techniques for many years. We will explore those techniques as well as recently popular algorithms like neural networks. In this post, we will focus on applying linear models on the features derived from market data.

StarSpace: Embed All The Things! (Paper Summary)

This paper describes a way to generate embeddings for various tasks. The algorithm is general enough which enables it to achieve strong results in very diverse tasks.

Deep Neural Networks for Youtube Recommendations (Paper Summary)

Youtube switched their recommender system from matrix factorization to neural networks few years ago. This paper describes the neural network models as well as the overall system around it, including the data processing and deployment aspects.

Deep learning networks for stock market analysis and prediction (Paper Summary)

In this paper, Deep learning techniques are applied to the financial market data directly rather than using any text/alternative data sources. This has been a relatively tricky dataset for any non-linear machine learning technique because of the extremely high noise-to-signal ratio. The authors use a relatively high-frequency dataset sampled at every 5 minutes. They consider 38 stocks from Korea KOSPI.

Deep Learning for Event-Driven Stock Prediction (Paper Summary)

In this post, I attempt to summarize this paper by Ding et al. This paper proposes a way to use the convolutional neural network on the news events for stock direction prediction. The data and the prediction granularity is 1-day.

Practical Text Classification for Production Systems

This post is about using a relatively simple yet powerful text classification model for a production text classificaiton system. Other topics like deployment, testing for out-of-sample texts are also discussed - they are often not the sexiest aspects, but it makes sense to discuss them in this post.

SELU vs RELU activation in simple NLP models

RELU activation function has become the de facto choice in neural networks these days. Few weeks ago, some researchers proposed Scaled Exponential Linear Unit (SELU) activation function. They show a far better convergence using SELU. In this post, I am posting a simple comparison of SELU against RELU using a simple BoW model on SNLI dataset.

Prophet - Time series prediction

Predicting daily (and intraday) volume is a classic time series problem in finance. We try to use the Prophet library for this task.