The Economic Cost of the Islamic Revolution and War for Iran: Synthetic Counterfactual Evidence

ABSTRACT

This study estimates the joint effect of a new political regime and war against Iraq, on Iran’s per capita Gross Domestic Product (‘GDP,’ constant 2010 US$) for the period 1978–1988, during the revolution/war. I use a synthetic control approach, whereby a synthetic Iran is constructed as a weighted average of other Middle East and North Africa (‘MENA’)/Organization of the Petroleum Exporting (‘OPEC’) countries to match the average level of some key per capita GDP correlates over the period 1970–1977 as well as the evolution of the actual Iranian per capita GDP during that period. I find a sizable negative effect of the joint treatment. The average Iranian lost an accumulated sum of approximately US$ 34,660 during 1978–1988 (i.e. the average annual real per capita income loss of US$ 3,150). This loss equals 40% of the real income per capita, which an Iranian could earn in the absence of revolution and war. The confidence sets based on constant, linear, and uniform assumptions of treatment effect show that estimated income loss for Iran is sizeable and statistically significant. The results remain robust to a set of placebo tests.

KEYWORDS:

• Synthetic control method
• treatment effect
• Iran
• Iraq
• war
• conflict
• revolution

JEL CLASSIFICATION:

• C23
• H56
• F51
• D74
• Q34

Acknowledgments

The constructive comments and suggestions of Editor (Christos Kollias), two anonymous referees, Vítor Possebom, participants in the SOAS Middle East Institute and the Centre for Iranian Studies Seminar (London, 2020) as well as conversations with Mehrdad Farahmand, Sven Fischer, Nima Mina, and Manouchehr Takin are greatly acknowledged.

Disclosure Statement

No potential conflict of interest was reported by the author.

Notes

1. Thus, I exclude countries such as Israel, Lebanon, and Iraq, which were experiencing conflict and violence during this period. The ‘Synthetic Iran,’ which is my counterfactual for Iran in the absence of the Islamic revolution and war with Iraq, is constructed using data during 1970–1977. The effect of the revolution and war is then estimated by comparing data for actual and synthetic Iran during 1978–1988.

2. While revising my study, I noticed a relevant research by Hasancebi (forthcoming) which examined the economic costs of the Iranian revolution, using SCM. My study differs from Hasancebi’s research in two significant directions: First, I consider a joint treatment of revolution and eight-year war with Iraq and estimate the economic costs for Iran. The war with Iraq was closely associated with the revolution and promotion of the ideology of the Islamic revolution by the Iranian government besides the presence of a fragile political system in Iran. Second, I have followed the recently developed methodology by Firpo and Possebom (2018) and Ferman, Pinto, and Possebom (2020) to gain more insights on the statistical significance of estimated income loss for Iran. By estimating confidence intervals around the calculated income loss (see Figure 2), I show that the calculated loss for Iran during Islamic revolution (1978) is not statistically significantly different from zero (compared to Synthetic Iran). However, the income loss gains both economic and statistical significance over post-revolution and the war with Iraq (1979–1988).

3. For a review of post-revolution economic development in Iran, see Farzanegan and Alaedini (2016).

4. See footnote 1 on the likely start of open resistance against the Shah in 1977. However, the intensity and violence of protests and repression were more visible in 1978.

5. In a standard panel data with fixed effects regression for my sample of 12 countries with 18 years (1970–1988), the included covariates explain approximately 65% within country variation in real GDP per capita. This figure increases to 86% after inclusion of lag of real GDP per capita in the right-hand side of the model.

6. I use Religion Adherence Data available at https://scholar.harvard.edu/barro/publications/religion-adherence-data.

7. http://www.worldvaluessurvey.org/WVSOnline.jsp.

8. Note that the difference between average values of co-variates of Iran (1) and unweighted average of co-variates for countries with weight > 0 (3) are not significant, showing that the selected countries in donor group (Table 1) are matching well with Iran in a term of socio-economic structure before 1978. However, there is a need to find optimum weights to assign to each of them to have the best prediction of outcome for Iran (real GDP per capita) before 1978.

9. On 23 October 1977, the suspicious death of Mostafa Khomeini, the son of Ayatollah Khomeini became the subject of a ‘martyrdom’ by opponents of Shah’s government and fueled the growing discontent with the Shah.

10. I re-examined the synthetic control estimation using 1977 as a treatment year (instead of 1978). The estimated impact of revolution and war remains largely similar to the original case (with treatment year 1978). This is not surprising since both 1977 and 78 refer to the revolutionary period which intensified in 1978 and collapsed the system in January/February 1979. The average annual income per capita loss for an average Iranian from 1977 to 1988 is, in this new estimation, US$ 3,193. This average annual income per capita loss is comparable to the earlier estimations of US$ 3,150 based on my original synthetic control estimations using 1978 as treatment year. Given these results, there is high level of confidence on the appropriate selection of treatment year (1977/1978). Using the results based on 1978 as treatment year is also justified based on inference tests (in-space placebo tests) which show a higher confidence in estimated impact compared to the case of estimation using 1977 as a treatment year. Finally, as is shown in Figure 2, the drop in income for Iran in 1977/78 is not statistically different from zero.

11. To check the inferences around the estimated average annual income loss (US$ 3,150) for Iran during the treatment period, I carried out the test of the null hypothesis that the mean of the income per capita gap between Iran and its synthetic, equals a specified value against the two-sided alternative that it is not equal to the specified value. This is based on a t-statistics test and the associated probability value is the p-value, or marginal significance level, against a two-sided alternative. If this probability value is less than the size of the test, (e.g. 0.10), I reject the null hypothesis that my estimated annual income loss per capita of US$ 3,150 equals to a specified value. The results from the Mean Test show that the estimated average annual income per capita loss can change from US$ 2,600 loss to US$ 3,700 annual loss per person (p-value > 0.10).

12. Note that confidence levels are discrete in a permutation test (such as the SCM’s placebo test) and depend on the sample size. Since I have 12 units, I can use confidence levels of 91.67% (which is 100–1/12*100) or 83.3% (which is 100–2/12*100).

13. We can see the same phenomenon in Figure 3 in Firpo and Possebom (2018). I appreciate the helpful discussions with Vítor Possebom on the topic of confidence sets.

14. In Figure 3, we can see countries with relatively higher (positive) changes after 1978 (compared to their pre-1978 levels). These are Malta and Oman. Malta has zero weight in building synthetic Iran and Oman has only 12% share. In sensitivity analysis, the influential countries are iteratively dropped from the donor pool to see how the estimated impact is robust to inclusion or exclusion of main countries in the donor group. As the Figure 5 shows, the overall picture is robust to this test. Furthermore, a clearer criterion is presented by comparing ratio between the post- and pre-intervention RMSPE in Table 6.

15. https://www.nytimes.com/1975/03/03/archives/shah-decrees-iran-a-oneparty-nation.html.

16. For more information on profile of countries see http://news.bbc.co.uk/2/hi/country_profiles/default.stm.