Capturing alternative data in practice: Transform news sentiment into actionable insights
Dr Axel Groß-Klußmann in Quoniam’s team Research Forecasts set out to find the best approach to analysing daily news sentiment scores for over 1,000 economic topics. The neural network models are competitive when it comes to explaining stock market movements, currency changes, bond yields and predicting economic growth. His work has recently been published in Digital Finance.
Axel, you recently published your research “Learning deep news sentiment representations for macro-finance” in Digital Finance. What motivated you to explore this topic?
At Quoniam, we have been working for over 10 years to understand whether sentiment coming from unstructured news flow data can help to predict future market movements. Undertaking the background research on this area, we utilised a range of methodologies to eventually come up with the most effective approach to choosing the news topics with the most market moving influence.
Whenever we have tried to bring together alternative sentiment data and financial time series, the biggest obstacle has always been the selection or weighting of the underlying narratives or topics. Some form of dimension reduction is required to link news collected on potentially thousands of topics to individual financial instruments. However, most of the standard statistical approaches left something to be desired. Often, too many data were discarded while other approaches produced weightings for topics irrespective of our use case context.
Given that neural networks are strong at these kinds of tasks, I was simply pursuing an idea I had about how to use neural networks to give us what we wanted. Recent software advances have made it very easy to customise and train neural networks, so it was just a question of seeing how the strengths of neural networks would carry over to our finance use case with considerably fewer data points than are typically used for training these models.
How did you approach these questions in detail?
So the key question we were trying to answer was which news topics have a market moving link? I tried using a range of statistical techniques like principal component analysis (PCA), partial least squares (PLS), ridge and lasso regression, to find patterns in the news data. These methods help reduce the complexity of data or prevent overfitting, but they can be limited when it comes to capturing very intricate or hidden relationships. To deal with the complex nature of news data, we needed more expressive models, which is why we turned to autoencoders – a type of neural network that excels at uncovering deeper patterns by compressing and reconstructing the data in a more flexible way.
Several software solutions in the autoencoder space are basically at your fingertips these days, which means you can implement a neural network in no time. Tech-wise, I employed PyTorch, a popular open-source deep learning package to crease a custom model called a supervised encoder. This model helps to both shrink data size and learn from labelled examples. Simply put, it helps the model understand and organise sentiment data in a way that fits our financial needs.
What were your findings?
On the plus side, training and coding the neural networks was easier than expected. Most importantly, the analysis showed that customising how we build statistical signals added significant value. Creating custom loss functions allowed us to tailor the model’s learning process, this gave us data organised in a manner that we liked much more than standard methods. Overall, neural networks consistently performed better in different cases. Interestingly, the depth or complexity of the networks was less important. While it was difficult to statistically outperform standard methods when accounting for multiple selection, the neural networks still provided data organised and transformed that were more useful for us as practitioners to then embed within our strategy analysis.
How do you apply these insights at Quoniam in the investment process?
The proposed approach is helpful in the research process itself. One example is the review of datasets containing text-derived numerical sentiment scores for many topics. The framework outlined in the paper gives us a very competitive benchmark approach that can directly produce actionable signals.
Additionally, we apply these insights in our liquid alternative strategy Quoniam Global Data Sentiment. Quoniam Global Data Sentiment is a liquid long/short strategy that identifies and acts on market-moving shifts in sentiment before they are priced in by analysing unstructured news flow to identify key investment signals. Our process starts with over 1 million stories split into over 50,000 topics captured every day. The neural network approach to dimension reduction helps us choose the most favourable of the 50,000 topics, giving us an edge in extracting the best return signals. The strategy has been live for over 3 years and has shown promising results.
Explore the paper:
Learning deep news sentiment representations for macro-finance
