The Bangko Sentral ng Pilipinas (BSP) recently announced how it explores applications of machine learning (ML) techniques, particularly in the areas of natural language processing, nowcasting, and banking supervision.
“Central banks’ interest in ML has been increasing over the years, mainly due to its potential to enhance the existing tools used for regular monitoring as well as its ability to uncover underlying relationships between data to better understand the economy and the financial system,” said BSP Governor Benjamin E. Diokno.
According to a study published by the Bank of International Settlements (BIS), some 80 percent of respondents from the 52 central banks surveyed discuss the topic of big data formally within their respective institutions. Meanwhile, 70 percent use them for economic research, and 40 percent use them to inform policy decisions.
“Central banks have substantial experience with large structured data sets, typically of a financial nature, but have only recently started to explore unstructured data,” the study explained, “Before they are analysed, they must be cleaned and curated, i.e., organised and integrated into existing structures.”
As for the Philippines, natural language processing at the BSP is used to convert text into data to produce a quantitative summary, such as the news sentiment index and the economic policy uncertainty index. Last January, the BSP announced that it will create the news sentiment index, a new high-frequency monitoring index for economic and financial-related news from online sources.
The BSP also works with ML approaches to generate nowcasts of regional inflation and domestic liquidity. These models supplement the central bank’s existing suite of models for macroeconomic forecasting.
In banking supervision, meanwhile, the BSP looks into utilizing ML techniques to enhance its data validation processes and better identify atypical data.
“While empirical studies provide evidence that ML techniques can outperform traditional regression analysis in forecasting, there is difficulty in interpreting the causal relationships in ML models. By contrast, traditional econometric models allow users to make inferences on causation and identify how an explanatory or independent variable influences the dependent variable. This is critical to economic analysis and policy formulation,” Diokno emphasized in citing the challenges presented by machine learning and artificial intelligence.
