ASEAN Central Bank Signal
ASEAN Central Bank Signal
From statements to market signals - Quantifying monetary policy communication
From statements to market signals - Quantifying monetary policy communication
Central bank communication plays an increasingly important role in shaping market expectations. However, interpreting monetary policy statements remains inherently subjective. Different analysts may reach different conclusions depending on their reading of the same document.
This framework provides a systematic and transparent approach to quantify changes in central bank communication.
The core question is: What policy signal does a central bank's monetary policy statement convey — hawkish, neutral, or dovish — and how does this communication evolve over time?
Rather than replacing traditional macroeconomic analysis, this framework is designed as a complementary analytical tool. It transforms qualitative central bank communication into measurable indicators that allow investors, economists, and market participants to track changes in policy tone, uncertainty, and thematic focus consistently across time.
The framework currently covers four ASEAN central banks:
Bank Negara Malaysia (BNM)
Bank Indonesia (BI)
Bank of Thailand (BOT)
Bangko Sentral ng Pilipinas (BSP)
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
The analysis is based exclusively on official monetary policy statements published by each central bank.
Each statement is treated as an individual document, allowing communication patterns to be measured across monetary policy meetings.
To ensure comparability across institutions and time periods, each document undergoes standardized text preprocessing:
Statements are collected from official central bank sources.
Document characteristics are calculated, including:
total word count;
number of sentences; and
average sentence length.
Text is converted into lowercase format.
Punctuation, symbols, and numerical values are removed.
Words are tokenized for textual analysis.
Common stop words and institution-specific boilerplate terms are removed.
The preprocessing stage ensures that the analysis captures economically meaningful changes in communication rather than differences caused by document formatting or repetitive institutional language.
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
The Hawk-Dove Index measures the overall monetary policy tone embedded within central bank communication.
A higher reading indicates a more hawkish communication stance, suggesting greater emphasis on inflation risks, policy normalization, or tighter financial conditions.
A lower reading indicates a more dovish stance, suggesting greater emphasis on growth support, downside risks, or accommodative policy conditions.
The index combines multiple established monetary policy and financial sentiment frameworks:
The LM dictionary identifies financial sentiment through:
positive language;
negative language; and
uncertainty-related language.
The framework is widely used in financial textual analysis because it is designed specifically for economic and corporate communication.
The ABG framework evaluates monetary policy communication at the sentence level. Unlike simple word counting, this approach considers whether policy-related sentences contain directional signals associated with:
tighter policy conditions;
easier policy conditions; or
forward guidance.
The BN framework provides a monetary-policy-specific hawkish and dovish vocabulary. It focuses on identifying language associated with:
inflation concerns;
policy tightening;
accommodation; and
downside risks.
The Correa et al. framework captures financial stability-related communication. This is particularly relevant for ASEAN central banks, where policy communication often incorporates:
financial system resilience;
credit conditions;
liquidity conditions; and
macroprudential considerations.
Generic dictionaries may not fully capture the communication style of ASEAN central banks. Therefore, a localized dictionary is developed separately for each institution.
The localized dictionary incorporates recurring policy-specific expressions identified from historical monetary policy statements. This allows the framework to capture institution-specific communication patterns while reducing the limitations of applying global dictionaries directly.
Each dictionary generates an individual sentiment score.
The scores are normalized by document length to account for differences in statement size.
LM = [Positive − Negative− (0.25×Uncertainty)]/Total Words
CEA = (Positive−Negative)/Total Words
ABG = (Hawkish−Dovish)/Total Words
BN = (Hawkish−Dovish)/Total Words
LSD = (Hawkish−Dovish)/Total Words
Individual sentiment dictionaries capture different dimensions of central bank communication.
For example:
LM captures broad financial sentiment;
ABG and BN capture monetary policy direction;
Correa captures financial stability considerations;
the localized dictionary captures institution-specific language.
To combine these measures into a single indicator, Principal Component Analysis (PCA) is applied.
PCA identifies the common underlying movement across the five sentiment measures and extracts the first principal component (PC1).
The resulting composite measure represents the dominant communication signal shared across different sentiment frameworks. The index is then rescaled between -100 and +100, where:
-100 to -34 :Dovish
-33 to +33 : Neutral
+34 to +100 : Hawkish.
A higher reading indicates a stronger hawkish communication bias, while a lower reading indicates a stronger dovish communication bias.
Important note: The index measures communication tone rather than the actual monetary policy decision. A central bank may keep rates unchanged while communicating a more hawkish or dovish outlook.
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
The Uncertainty Index measures the degree of uncertainty expressed by policymakers.
Higher uncertainty readings indicate greater emphasis on:
risks;
downside scenarios;
uncertainty around economic conditions;
external developments; or
policy challenges.
The index is based on the uncertainty dictionary developed by Loughran and McDonald (2011).
First, uncertainty intensity is calculated as:
Uncertainty Rate = Number of Uncertainty Words*1000/Total Words
The measure is then converted into a 0–100 historical index:
Uncertainty Index = (Current − Historical Minimum)*1000/Historical Maximum−Historical Minimum
Interpretation:
0–30: Low uncertainty
30–60: Moderate uncertainty
60–100: High uncertainty
The calibration range is fixed based on historical observations to ensure consistency when new statements are added.
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Central banks discuss multiple economic issues simultaneously. Topic Evolution identifies which themes dominate communication and how their importance changes over time.
The analysis uses Latent Dirichlet Allocation (LDA), an unsupervised machine learning technique introduced by Blei, Ng, and Jordan (2003).
LDA assumes that each document contains a mixture of underlying topics, while each topic consists of a probability distribution of words.
The model identifies recurring combinations of words that frequently appear together across monetary policy statements.
Examples of themes identified include:
inflation and price pressures;
external sector developments;
domestic demand;
financial stability;
credit conditions;
growth outlook;
liquidity conditions.
The number of topics (k) is selected independently for each central bank based on:
statement length;
historical coverage;
communication complexity; and
topic interpretability.
Because monetary policy statements differ significantly across institutions, applying a fixed number of topics across all central banks may reduce analytical quality.
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
The Top Words/Phrases section highlights frequently mentioned terms in recent monetary policy statements. The purpose is to provide a quick market-oriented summary of communication emphasis. The analysis extracts:
frequently occurring words;
frequently occurring phrases;
recurring policy expressions.
The extraction process uses:
unigram analysis (single words);
bigram analysis (two-word phrases).
Institution-specific terminology and generic monetary policy language are removed to ensure that the remaining phrases represent meaningful changes in communication focus.
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
This framework provides several advantages:
The same methodology is applied across multiple ASEAN central banks, allowing comparison over time.
Each indicator is derived from observable textual features rather than subjective analyst interpretation.
The framework combines:
established sentiment dictionaries;
monetary policy-specific dictionaries;
institution-specific language;
machine learning-based topic modelling.
This reduces reliance on any single measurement approach.
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Text-based measures should be interpreted as complementary indicators rather than substitutes for economic judgment.
Several limitations remain:
Words may have different meanings depending on context.
For example, discussion of “higher inflation” may indicate either concern about inflation persistence or merely describe past developments.
Central bank communication often contains deliberate ambiguity. A quantitative index may not fully capture subtle forward guidance.
Dictionary-based methods depend on predefined terminology and may miss newly emerging expressions.
The indices measure changes relative to historical communication patterns. A high or low reading indicates stronger or weaker communication intensity compared with the past, rather than an absolute judgement of policy appropriateness.
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
This framework provides a systematic approach to analyzing ASEAN central bank communication by transforming monetary policy statements into measurable indicators of:
policy tone (Hawk-Dove Index);
communication uncertainty (Uncertainty Index);
dominant policy themes (Topic Evolution); and
recent communication emphasis (Top Words and Phrases).
By combining textual analysis, sentiment measurement, and machine learning techniques, the framework aims to provide investors and economists with a consistent lens to monitor how central bank communication evolves through different economic cycles.
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Apel, M., & Blix Grimaldi, M. (2014). How informative are central bank minutes? Review of Economics, 65(1), 53–76.
Bennani, H., & Neuenkirch, M. (2017). The (home) bias of European central bankers: New evidence based on speeches. Applied Economics, 49(11), 1114–1131.
Blei, D. M., Ng, A. Y., & Jordan, M. I. (2003). Latent Dirichlet Allocation. Journal of Machine Learning Research, 3, 993–1022.
Correa, R., Garud, K., Londono, J. M., & Mislang, N. (2021). Sentiment in central banks’ financial stability reports. Review of Finance, 25(1), 85–120.
Loughran, T., & McDonald, B. (2011). When is a liability not a liability? Textual analysis, dictionaries, and 10-Ks. The Journal of Finance, 66(1), 35–65.