Time travel has always been a crazy notion. But in a way, it’s what David Ardia does. Along with his team of researchers, this associate professor in the Department of Decision Sciences at HEC Montréal is exploring the feelings of Quebecers of decades past, and even those of the future, using stacks of old newspapers (yes, time machines can come in all shapes and sizes!). By combining econometrics with artificial intelligence, the team is attempting to find out whether our ancestors of the past century were worried, confident or skeptical about the economic policies of their time.
To answer those questions, since he can’t actually travel backward in time, Professor Ardia relies on sentometrics—that is, the study of sentiment embedded in texts. Combing through recent as well as older writings, he locates polarizing words, whether positive or negative, and converts them into quantitative data. This methodology enables him to analyze texts and track the evolution of feelings expressed over the years.
Before they could unravel texts to go back in time, Professor Ardia and his team had to take a deep dive into the archives. In collaboration with CBC/Radio-Canada and the CO.SHS project, led by Bibliothèque et Archives nationales du Québec (BAnQ) and funded by the Canada Foundation for Innovation, they consulted news stories from four French-language media outlets in Québec, La Presse, Le Soleil, Le Devoir and Radio-Canada, going back to 1913. As the researcher explains, that rich trove of documentation sets this project apart from similar initiatives: “The sentometrics-based economic policy uncertainty index established for the rest of Canada relies on news stories dating back to 1985 only, and is limited to English-language media.”
Once he had all the articles, Professor Ardia set about building three dictionaries: “I assembled a lexicon of words with connotations of uncertainty, like crise, anxiété and alarmant [‘crisis’, ‘anxiety’ and ‘alarming’]. Then I repeated the operation for the economy and policy.” Once the dictionaries are compiled, artificial intelligence comes into play: algorithms are used to detect the dictionary words appearing in the articles. If a text contains at least one word from each of the lexicons, the researchers know that economic policy uncertainty (EPU) was in the news that day.
Understanding old newspapers
Locating these triplets of keywords in recent news stories is child’s play for the algorithms. Things get more complicated when studying texts from the past, when most newspapers printed daily editions. On those pages, articles were laid out in columns and printed one after the other, which makes decoding the words substantially more complex. “Imagine that you have words relating to policy in one article, and a bit further on in the same column, there’s a sports story that contains a word that signals uncertainty,” Professor Ardia explains. “This would be incorrectly flagged as EPU.”
To ensure sentometrics works with older newspaper editions, Professor Ardia’s team developed algorithms that isolate the positions of the words in a triplet to determine whether they are in close proximity to each other. Essentially, they find the words, then locate them “geographically” within the text to find out if they are in neighbouring lines.
Tied to historical events
To compile this astronomically large quantity of information into a more digestible format, Prof Ardia has created an index that calculates the number of occurrences of triplets. The more triplets are found, the higher the EPU index. Consequently, the index was high during key periods when socio-economic upheavals affected Quebecers’ daily lives. These include the 1918 influenza pandemic, the Great Depression, the Oka Crisis and the financial crisis of 2008–09.
To make sure the index properly reflects reality, the researcher applied dynamic standardization; in other words, he weighted the media sources: “If you have some media that are strongly focused on economics and policy, this can artificially drive the index higher. A similar problem arises if a newspaper specializing in economics stops publishing at some point. Standardizing the index prevents those biases from influencing the number of occurrences of triplets.”
Travelling into the future
But why spend so much time and effort retracing the emotions of people in the past?
It allows us to predict the future: with our index, we can do short-term macroeconomic forecasting.
To make sure the index is a better predictor than a crystal ball, he conducted a number of historical verifications: “I took dates at random, then jumped forward a month in time to see whether the macroeconomic forecast worked. For example, I might choose February 1980, make a forecast, and compare it with what really happened in March 1980.”
The index’s reliability makes it a valuable tool for decision-makers, and it has many applications. There is a correlation between an increase in the value of EPU and reduced activity by domestic companies in the ensuing months. Therefore, if it observes a spike in the index, a government can anticipate a slowdown and stimulate economic activity by introducing a business support program. The index is also useful to private investors, who can strengthen their positions in financial markets when the index is low, which is a signal of stable economic policy.
To go further
Reference of the published article