Some data and tech applications that can improve central bank activities include extraction of quantitative information from texts and images, which can be useful for forecasting financial market variables, predicting headline inflation, and analyzing the real estate market, said Daniele Franco, deputy governor of the Bank of Italy, in closing remarks during a 2-day workshop.
In addition, machine learning algorithms can be used for combating illicit financial activities and increasing the statistical accuracy of micro- and macro-economic indicators (for example, forecasts of the Index of Industrial Production).
Some indicators have moved rapidly from the experimental to the production stage. For example, Banca d’Italia has developed and now regularly uses an indicator of the situation of the main banks obtained using selected tweets. In addition, the central bank has improved its estimates of Foreign Direct Investment for the Balance of Payments by introducing Machine Learning classifiers.
Also, they are working on other sources of information, such as credit and debit card payment data, which should help assess consumption, retail sales and other macroeconomic variables.
These developments present significant challenges.
First, there is a risk of biased statistical inference. Big data and, in particular, unstructured data collected from social media may reflect only a part of the population. Increasing the sample does not improve the accuracy of the estimates if the source of the bias is not understood and dealt with.
The use of massive amounts of rough data may result in an exceptionally good in-sample fit but may perform poorly out-of-sample (the curse of overfitting). Moreover, when using big data, special attention should be paid to the robustness and stability of the estimates over time.
The second challenge is privacy. Protecting the integrity and confidentiality of personal data is of paramount importance to our societies. Storing personal data in digital repositories belonging to public and private institutions can be risky. We should avoid any use for illegitimate purposes such as the second-hand trading of personal information.
Banca d’Italia is exploring new avenues for utilizing microdata without jeopardizing their confidentiality. In this way data from different institutions can be combined to produce aggregate statistics that are important for understanding economic, financial and welfare issues.
They have started experimenting with some privacy-preserving algorithms from Istat (Italian National Institute of Statistics), and are also discussing these issues with Eurostat (statistical office of the European Union) and the National Authority for Personal Data Protection.
The third challenge concerns the preservation of data over time. While central banks keep producing a huge amount of data, there is not yet agreement on a shared set of criteria for making these data accessible in the future. It is crucial that technology-neutral standards be established to ensure data availability for future generations. Coordination between statistical agencies, governments, academia and private companies is equally crucial.
Timely and accurate information derived from non-traditional data sources, as well as new analytical techniques can help central banks to improve their knowledge of the economy and society, allowing them to make better-informed data-driven decisions.