The Conduct Chronicles – “I Spy with my Little AI”

Many of us have multiple social media accounts – perhaps one for work, another for friends and family, and perhaps a third that we use to share our thoughts on breaking news, current affairs and the reasons behind the soaring price of butter. Yet when we think about social media, and our network of followers, we are unlikely to consider that regulators, or indeed central banks, might be taking a keen interest in our posts. However, that is exactly what is happening. But why?

Thankfully, we don’t need to go far to gain some insights as, in most cases, we only need to read their online policy statements.

The European Central Bank’s (ECB) social media monitoring policy explains its curiosity, advising that it tracks different online sources including forums, blogs and online news websites to ‘understand how social media users discuss monetary policy and banking supervision, as well as other topics relevant to the ECB.’ The ultimate aim? To help shape the ECB’s communications strategy.

Meanwhile, in the UK, the FCA has also set out its approach to its research, advising that it collects personal data ‘to get a holistic and accurate view of the markets, understand consumer behavior and properly identify issues, trends and risks.’ Part of that data collection includes social media (eg. X (formerly Twitter), Facebook, blogs and news articles).

Rise of the Machines?

The fact that these organizations are taking a keen interest in social media is unsurprising perhaps given its pervasiveness across all sectors of society. However, what may be more surprising is the extent to which central banks and regulators are exploring AI to expand their capabilities, and the different use cases that are emerging.

In Oct 2021, central banks met to discuss the application of machine learning (ML) in the increasingly complex environments in which they operate. In amongst the expected discussions around data gathering, delegates noted that ML can enhance data quality, for example, by ‘dealing with outliers, addressing the problems posed by missing values, limited frequency and/or timeliness, and by providing richer contextual insights.’

Also discussed, was the marked increase in the use of ML to enable SupTech, that is, the use of new technologies and big data analytics to support supervision. These technologies can help drive greater efficiencies, facilitate the assessment of micro-level fragilities, and help identify emerging topics and risks.

However, there were some far more unexpected scenarios discussed, including insights from Banque de France which outlined a study where it harnessed non-traditional indicators from social media networks to estimate inflation perceptions. Its study, based on data from Twitter (now X), focused on accounts that retweeted posts of the Banque de France account. It classified all tweets using a combination of ML, natural language processing along with dictionary-based filters to create ‘a Twitter indicator of inflation perception.’ The result? The study highlighted that the indicator was ‘consistent with monthly household surveys on inflation expectations and perception, and is highly correlated with the inflation rate.’

Meanwhile, research think tank Official Monetary and Financial Institutions Forum (OMFIF) has also set out the ways in which various central banks have been testing with, and deploying, AI including:

1. The Bank of Indonesia using ML to incorporate the impact of foreign investor behavior on exchange rates and monetary policy decisions,
2. The Bank of England and ECB using AI to monitor data quality for signals of unexpected economic shocks, and
3. The Central Bank of Malaysia applying AI to newspapers (approximately 750,000 articles) to improve its forecasting accuracy of gross domestic product growth.

These examples prove that a powerful AI system programed with well-defined parameters can be used to turbo-charge the reading and transformation of text into data usable as part of a more holistic picture of the economy.

The Trouble with Data

Notwithstanding the innovative ways in which central banks and regulators are using AI, including capturing and analyzing comments from social media, the trend undoubtedly poses concerns and questions around data integrity, bias and privacy.

Given the sheer amount of personal information available via social media platforms, and the active monitoring and analyzing underway, is it sufficient that they point to the relevant laws stating that they adhere to them? Even with data regulations in place, given some rather spectacular data breaches over the past few years, some individuals may feel rightly concerned about the security and protection of any personal-level data being held and the extent to which it is sufficiently anonymized and protected.

Regarding potential bias, the Banque de France study notes that it analysed only those Twitter (now X) posts that were a ‘re-Tweet’ from their account. Has this introduced bias into the equation given other voices have been excluded?

More broadly, what this trend highlights is a need for these organizations to continue to invest in data scientists with the skills to recognize and address the challenges and risks, and critically, to ensure that AI guardrails are implemented and followed.