An Epidemiological Model of Economic Crisis Spread across Sectors in the United States

Abstract

This paper develops a discrete‐time epidemiological model to characterize the spread of economic deterioration across sectors in the United States for the period 1952–2015. It is the first model to apply an epidemiological approach to consider such spread using macroeconomic Flow of Funds data. By extending the usual one‐period Markov model to a two‐period setting, we incorporate the possibility that an initial slow growth period may either continue further or improve such that further economic deterioration is averted. The estimated model can be used to classify more versus less contagious sectors and identify their channels of transmission.

Sectoral slowdowns in the United Kingdom: Evidence from transmission probabilities and economic linkages

“…One way of modeling cross-sectoral spillovers is via the two-level epidemiological model with immediate recovery, used in Janssens et al (2021). 7 A short description of this model is given in this subsection; more details are provided in Appendix A in the Supporting Information.…”

“…Note that this probability is computed as one minus the probability of not being contaminated by any of the potential sources of contamination, these sources being nature (exogenous contamination) or any of the infectious sectors in the set Infectious t−1 . As shown in Janssens et al (2021), the first and second set of parameters (the recovery probabilities) can be estimated independently of the other parameters in the model. 9 From the estimated parameters, the expected number of sectors that a sector will contaminate in an otherwise susceptible system can be computed; this quantity is called the basic reproduction number in the epidemiological literature and is denoted by R 0 .…”

“…The Bayesian approach also mitigates another drawback of the statistical epidemiological approach, encountered in Janssens et al (2021): that a large number of parameters need to be estimated, increasing quadratically in the number of sectors. Depending on the time span for which the data employed is available, some of these parameters may not be identified or are estimated on the basis of very few events.…”

“…There is nothing special about our choice of this model as a baseline to represent the traditional epidemiological approach. Any Markov-based epidemiological model can be used, but this section follows the notation and assumptions of the specific model mentioned.8 We adopt the mnemonics inJanssens et al (2021), which refer to how these states might be interpreted in a country setting: an initial slow growth state could be seen as a downturn (D), whereas protracted slow growth can be interpreted as a country being in an economic crisis (C). As discussed inJanssens et al (2021), other thresholds can also be adopted to determine the states a sector is in.…”

“…Any Markov-based epidemiological model can be used, but this section follows the notation and assumptions of the specific model mentioned.8 We adopt the mnemonics inJanssens et al (2021), which refer to how these states might be interpreted in a country setting: an initial slow growth state could be seen as a downturn (D), whereas protracted slow growth can be interpreted as a country being in an economic crisis (C). As discussed inJanssens et al (2021), other thresholds can also be adopted to determine the states a sector is in. The decision to consider episodes of one or two quarters of negative net worth growth comes from the analogy to the macroeconomic literature, where two consecutive quarters of negative GDP growth conventionally signals a recession.9 The model used in this paper does not allow for recovery of sectors to depend on the states of the other sectors in the economy.…”