Enhancing Central Bank communication through Natural Language Processing: Mexico and the IMF

This paper derives principles for maximizing verbatim replication of key central bank messages by applying common Natural Language Processing (NLP) techniques. In particular, we lay out an application of the n-gram bag-of-words model to inform central banks' communication strategies involving the Flexible Credit Line (FCL), a crisis-prevention and crisis-mitigation instrument that grants immediate access to liquidity during contingencies and requires no ex-post conditionality. The FCL provides an interesting case study since it has the explicit objective of reducing the perceived financial and political stigma of borrowing from the IMF.

Central banks that subscribe renewable FCL arrangements face a three-fold communication challenge: signaling their short-term liquidity management strategy clearly, detaching from the perception associated with conditionality of traditional IMF programs, and explaining FCL characteristics and limitations to the general public. Considering these, we first study clarity, readability, and syntax of central bank FCL-related press releases. Then, we estimate text-based similarity and containment measures between official statements and corresponding FCL media coverage in newspapers and written articles online. Lastly, we asses transmission and replication of key messages. Our main focus is on Mexico; however, our empirical strategy is applicable to central banks whose communication is primarily carried out in Spanish and so we also asses FCL official communication in Colombia and Chile–albeit with much smaller news samples.