Chennai 2015: A novel approach to measuring the impact of a natural disaster

by Ila Patnaik, Renuka Sane, Ajay Shah.

In November and December 2015, the city of Chennai in the Southern Indian state of Tamil Nadu, got heavily flooded owing to unprecedented rainfall. With a population of a little more than 7.1 million people, Chennai is one of the major urban centers of South India, and one of the four important metropolitan cities in India. The flooding is estimated to have led to the loss of more than 500 lives, and damages of about US $3 billion, making it the world's eighth most expensive natural disaster in 2015. In this paper we evaluate the impact of this event for households in Chennai.

Natural disasters, such as the Chennai floods, are important shocks which can influence all parts of the income distribution. In the aftermath of such a natural disaster, the issues of consumption smoothing, liquidity constraints and financial resilience play out. Natural disasters are important in their own right, as we need to understand more about the turmoil faced by households in such states of nature. All governments engage in redistribution in the aftermath of a natural disaster. This motivates research on studying the impacts of natural disasters. Natural disasters are also an opportunity to obtain insights into the economics of household, through observation of households when confronted with such a large shock.

Many researchers have gone into the field after a natural disaster has taken place, and produced evidence about health, income, consumption, and financial conditions in the aftermath of the disaster. But such research does not offer insights into the causal impact of the event as adequate information gathering about baseline conditions, before the event, is lacking.

When panel data about households is present, we observe households before and after the natural disaster. This makes possible the analysis of the adverse impact upon affected households, while additionally observing controls. The constraint in such research has been the time elapsed between two consecutive observations of each household. As an example, even if a panel is measured once a year, there would be many months of elapsed time between the two measurement dates that bracket a disaster event.

In a new NIPFP working paper, Chennai 2015: A novel approach to measuring the impact of a natural disaster we exploit the new opportunities for measurement which flow from the CMIE Consumer Pyramids Household Survey ("CPHS"), which measures a panel of 170,000 households across India. Each household is met with three times a year. There is thus a period of four months, across which the household is measured twice, within which each natural disaster lies. We setup difference-in-difference estimation where households in Chennai are the ``treatment'' group and unaffected households in the rest of the state of Tamil Nadu are the ``control'' group. As households in Chennai are among the more affluent ones in Tamil Nadu, the raw dataset has poor match balance, and we address this problem by also performing matched DiD analysis.

We investigate three questions. First, we evaluate the impact of a flood on household income and consumption expenditure. It is possible that a disaster leads to declines in household income and expenditures owing to the destruction. However, it is also possible that households increase their spending to cope with the disaster, or replace capital stock. For example, some household activities, such as cooking, would shift from internal production to purchases from external providers, which would augment demand for certain goods and services. Households would start buying goods and services for reconstruction almost immediately after the destruction. Large scale expenditures on relief and reconstruction by the Indian state would bolster the local economy.

We find that there was no statistically significant impact on household income during the flood months. Households in Chennai, however, saw a 32% increase in consumption expenditure relative to the non-affected districts. The largest percentage increases in expenditure were seen on health, and power and fuel.

A key figure is shown above. The dotted line is for the controls and the deep green line is for households in Chennai. In both cases, what is shown is the monthly expenditure per person. The vertical black lines bracket the flood events.

At the outset, the households observed in Chennai are, on average, more affluent than the controls. Roughly speaking, we do have parallel trends in the period prior to the flood. During the flood, there was a large surge in expenditure which runs for many months. After that, consumption went down, to a point where the Chennai households were now comparable with those seen in the rest of Tamil Nadu.

Second, we evaluate the variation in the change in expenditure for different households. The adverse impact upon persons who live in structures with inferior structural strength is likely to be larger. We categorise households as more vulnerable, or more financially constrained, through various characteristics such as not having a concrete roof, or not having modern finance (such as life insurance, mutual funds, equity market participation), or not having durable goods (such as ACs, refrigerators etc). We find that the consumption expenditure of the these weaker households increases by a smaller amount than those not financially constrained. This might mean more hardship, and a higher inability to cope with catastrophic events.

Third, we evaluate the mechanism that households use to finance the higher consumption. Households could either draw down their savings, or increase their borrowings to finance expenditures. Our analysis suggests that relative to the control group, fewer households in Chennai saved, borrowed, or purchased assets, in the period after the floods. This suggests that reduced savings and reduced purchase of assets was the channel through which the consumption surge was financed. In our data, after about a year, the consumption surge ended, and was followed by a further decline in consumption. This may be consistent with households refocusing on repairing their balance sheet.

Natural disasters kill around 90,000 people and affect close to 160 million people worldwide. The frequency and intensity of disasters are expected to increase with global warming. Greater understanding is required about how natural disasters impact economic outcomes, so that better public and private responses may be designed. The contribution of this paper lies in bringing new tools of measurement (panel data, three times a year, matched DiD) to bear on an important problem (natural disasters) and discover the phenomena that are at work. The novel estimation strategy shown here can now be applied for many natural disasters in India. Over time, a body of work can develop of this nature, through which more abstract insights can be obtained.



The authors are researchers at NIPFP.