24 May, 2021

Why another business cycle indicator?

  • Economic variables and indicators are published with significant delay
  • Difficult to assess the health of the economy during the rapidly evolving Covid–19 crisis

Why another business cycle indicator?

  • We need high-frequency information to assess how stricter or looser health restrictions affect the economy
  • Our contribution: A novel daily fever curve (f-curve) for the Swiss economy using publicly available financial market and news data

Components of f-curve

Financial Market Data

  • Risk Premia: Difference between corporate and confederaton bond interest rates
  • Uncertainty: Stock market volatility
  • Term Spread: Difference between long-term and short-term interest rates
  • Attractiveness of the Swiss franc: Difference between short-term interest rates in Switzerland and abroad

Components of f-curve

News Sentiment

  1. Search for articles in online archives (keywords: business cycle, recession, economy)
  2. Download lead texts and titles of these articles
  3. Count positive and negative words using a lexicon.
  4. Sentiment = (#P - #N)/#T
  5. Calculate average sentiment of all articles

Estimation of f-curve

  • Estimate a factor model in static form: \[ X= F \Lambda + e \]

with \(X\) = data matrix, \(F\) = common factors. \(\Lambda\) = factor loadings, \(e\) = error term - f-curve is the first principal component of the static factor model

Advantages:

  • Parsimoniously summarize the information
  • Clean the indicators from idiosyncratic fluctuations

f-curve since 2000

  • Indicator closely tracks economic crises

  • The peak during the Covid-19 crisis is comparable with the Financial Crisis. But the speed of the downturn is considerably higher

  • The Covid-19 crisis is less persistent

f-curve during Covid-19 crisis

  • Begins to increase in February

  • Reaches the peak in March, just after national lockdown

  • Up to 16 June 2020, the f-curve improved to 1/4 of its peak value during the lockdown

In-sample Analysis

Out-of-sample Analysis

  • Pseudo out-of-sample realtime evaluation using direct forecast model: \(Y_{T+h} = \alpha_h + \beta_{h,1} f_{T|t} + \beta_{h,2} f_{T-1} + v_{T+h}\)

  • f-curve is not significantly worse than SECO‘s first estimate but available 2 months earlier

News vs. Financial Indicator

  • Financial market data respond more strongly during crises

  • News Indicator is much more volatile

  • Much room for improvement for the news indicator

Future Avenues

  1. News sentiments could exploit more sources, especially from the French- and Italian-speaking parts of Switzerland

  2. Use a topic modeling algorithm, instead of own search queries, to classify news

  3. Lexicon could be tailored specifically to economic news

Conclusion

  • Develop a daily indicator of Swiss economic activity using financial market and news data

  • Evaluation shows that it‘s correlated with other business cycle indicators and accurately tracks Swiss GDP growth

  • Major strength: can be calculated with a delay of 1 day

Appendix

Revisions

Decomposition