- DNB has unleashed complex algorithms on nearly a million articles that have appeared in FD over the past 36 years.
- The result: an indicator that compares the general feeling in the newspaper to the state of the economy.
- This gives a faster and more up-to-date picture of economic growth than commonly used measures such as those of the Central Bureau of Statistics.
“ Beatrix hospital closed due to coronavirus: patient tenth confirmed, ” headline Het Financieele Dagblad on March 1, 2020. In the days, weeks and months that followed, articles pile up about the coronavirus outbreak: ‘Nearly twenty infections in the Netherlands’ and ‘Roti turns into crisis mode’ and ‘More and more European Union countries are asking about the distribution of Corona family members.’
The newspaper spoke of the enormous uncertainty and the great economic consequences. This gave the reader a real-time dashboard of the free fall of the Dutch economy.
This isn’t an empty marketing slogan for FD makers, but a scientific fact, according to research by Jasper de Winter and Dorinth van Dijk of De Nederlandsche Bank (DNB). They have unleashed complex algorithms for nearly a million articles that have appeared in FD over the past 36 years. And with that, they created an index comparing feelings in the newspaper to the state of the economy. That relationship turned out to be remarkably close. “The sentiment index we have developed parallels the Dutch economy.”
What exactly is under the hood of their arithmetic model? “Our algorithms in FD look for words that indicate feelings, such as“ pull back, ”“ loss, ”or“ recover. ”The model is divided into four main categories: financial markets, business, economics, and politics. That way we cannot determine the degree of sentiment for each article, de Winter says. Not only can we find out exactly where it comes from.
Bringing these different degrees together produces a measure of the Dutch economy. Because there is a new newspaper nearly every day, this index provides a faster and more up-to-date picture of economic growth than commonly used metrics such as those of the Central Bureau of Statistics. Statistics Netherlands does not provide figures on GDP development until 45 days after the end of the quarter. Moreover, these are tentative numbers that are often revised at a later time.
Although they also looked at data from other newspapers, according to the researchers, it was clear working with articles from FD. “We wanted to focus on local financial economic news, and that’s exactly what you are writing about.”
Almost all partitions can be used
This allowed them to use almost all sections of the newspaper. Only the opinion column and the one-time English page were included in the newspaper, as they would only be a distraction. They should have skipped more articles from other newspapers. If we had taken a “regular” newspaper, a large number of articles would have been irrelevant to our investigation. This will create additional noise in the data, Van Dyck says.
The halo crunch has well illustrated the spot value of the FD. The fact that the epidemic caused an unparalleled shock to the Dutch economy in March of last year is clearly visible from almost vertical downlines. Uncertainty and fear explode pages, and are also captured by the DNB account model. This is self-learning, which means it quickly considers words like “Covid” and “Corona” closely related to the Dutch economy. Van Dijk explains: “If these words appear too often, they will automatically be included at a certain point.” “It didn’t happen on the first day, but you can see the algorithm picks up the crisis in time.”
De Winter adds: “It is not a positive thing that feelings become very negative because of Corona, but as researchers we are of course happy with that, because it shows that our model works.”
Also remarkably, negative sentiments in March appeared in all sub-categories: in politics and the business world, but also in articles related to financial markets and economics. He explains that the Corona crisis has afflicted the entire society. According to De Winter, this clearly differs from the position of the index surrounding the financial crisis that erupted in full force in September 2008. De Winter explains: “As of 2007, the period preceding that crisis can be seen mainly in the negative mood surrounding the financial markets only later. In 2008 and 2009, the change in sentiment will also be visible in the other categories. ”
It is reassuring for any economic-financial newspaper that its tone is not harmonious and that it is on the pulse of the economy. But it goes further, de Winter and Van Dyck noted. “The FD Sentiment Index can also forecast the direction of the Dutch economy well.”
The idea of making economic forecasts based on news sentiment is not new. In the early 1990s, the British weekly The Economist counted the number of times the word “recession” appeared in high-quality British newspapers every day. It turns out to be a good indicator of the coming recessions in the UK. It turns out that this is also the case for the famous column Keep up with the market In the Wall Street Journal. Its content appears to anticipate today’s moves in the New York stock markets.
Professional investors and stock traders also use algorithms that scan headlines. By responding to company news faster than the rest, they try to anticipate price movements. For example, Dutch company Semlab offers dozens of sentiments about stocks to clients of Deutsche Börse and JPMorgan’s investment bankers, among others. The points vary between plus and minus ten, with the number 10 being a buy signal. Minus ten means selling.
We have two quarters
However, DNB researchers do not focus on today’s issues. They try to look ahead a quarter to two quarters. To that end, they want to combine news sentiment with the tools DNB is already using for this purpose. But first, the gentlemen continue their research. In the near future, they will document the proof of their algorithms and their predictive value in a scientific publication. The intent is that DNB will also use the indicator in practice.
FD also wants to use the indicator. It is not clear how many times it was published. The editor says it would be a shame to do nothing with this valuable wealth of data. Especially now that this value has been scientifically proven.
Researchers are also satisfied with the results so far. De Winter: The results for the Netherlands are in line with the results of previous research abroad. Newspaper reports are an important source for monitoring and forecasting the economy.
“Coffee fanatic. Friendly zombie aficionado. Devoted pop culture practitioner. Evil travel advocate. Typical organizer.”