The words central banks use to explain policy decisions matter. They can, in some cases, affect financial markets just like changes in policy interest rates do. For this reason, we built a tool to track the tone of the Bank of Canada’s policy communications and assess how tone affects market perceptions.

Central banks influence interest rates on mortgages and business loans by changing their policy interest rate. But the policy rate isn’t the only item in their tool kit. Communications with the public and with market participants are also important because they can shape expectations about the economy and inflation.

Market participants pay close attention to these communications, including those from the Bank of Canada, to get a sense of possible changes in the central bank’s outlook or policy stance.

In turn, the Bank tracks how markets interpret those messages. This allows the Bank to see if its messages are being perceived differently than intended and to identify ways to improve its communications, if necessary.

The tone used in communications can affect how messages are understood and interpreted. Measuring tone consistently is challenging because it must be inferred from the wording of the text, and because of the frequency and large volume of Bank communications. To address these issues, we use advanced machine‑learning methods to measure the tone of Bank communications. We also compare the tone of the Bank’s messages with the tone used in commentaries by economists in the financial sector. What we find is that the tone of these commentaries typically shifts toward the tone the Bank uses when communicating its decisions on the policy interest rate.

We analyze words to measure tone

Monetary policy decisions are complex and are shaped by economic factors such as inflation, labour market conditions and economic growth. The Bank’s communications related to these decisions contain signals about all these elements. Together, these signals establish the tone of the Bank’s communications.

In particular, monetary policy announcements use wording that can be put into three categories:

• dovish wording suggests an easing of monetary policy
• neutral wording suggests neither an easing nor a tightening of monetary policy
• hawkish wording suggests a tightening of monetary policy

To calculate tone, we use a large language model that we fine-tune with examples of the three types of wording. The model reads a document and classifies each sentence as either dovish, neutral or hawkish. It then gives each sentence a value:

• dovish sentences receive a value of -1
• neutral sentences receive a value of 0
• hawkish sentences receive a value of +1

We then calculate the average of all sentence values in the document to obtain a single tone score that ranges from -1 to +1 (Figure 1).