CGE assumptions under scrutiny: uncertainty in the ethanol promotion policy in Mexico

ABSTRACT

Based on a CGE exercise of a subsidy to initiate ethanol production in Mexico, we use Monte Carlo simulations for consumer demand elasticities and ethanol cost estimates. The analysis provides three conclusions: when markets vary smoothly and predictably, Monte Carlo methods can then help to gauge the actual probability that a given program will achieve a desired outcome. Second, secondary markets may display little or no sensitivity to these parameter variations. Finally, a ‘razor’s edge’ outcome with no positive benefits if a critical parameter falls below some critical value, reveals that an economic policy may not be conducive to ‘fine tuning’ by marginal adjustments.

KEYWORDS:

• CGE
• Monte Carlo
• ethanol
• Mexico
• MPSGE

JEL CLASSIFICATION:

• C15
• C68
• D58
• D81

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ In one of the tendering processes a winner was, in fact, announced. The winner, however, declined the contract citing recent increases in sugar cane input costs.

² Government officials considered that a mandatory policy on MTBE substitution could endanger gasoline supply, if ethanol supply did not meet their targets. Thus, the policy guidelines changed in their posture towards fuel substitution and began to recognize the role of ethanol as an additive.

³ Data was extracted from the input-output Brazilian matrix. Although not explicitly stated in the IO matrix (since it is embedded in the agricultural inputs), in Brazil sugarcane bagasse is used as energy source during processing in Brazilian mills (as explained in Claros Garcia and von Sperling 2017). We assume Mexican producers would adopt the same practice.

⁴ Our ethanol cost estimates were obtained from the Brazilian IO matrix, and adapted utilizing local information to fit production and cost conditions in Mexico.

⁵ Using error corrections models Galindo (2005) has carried out an exhaustive study of energy markets in Mexico and we base our initial elasticity value in our CGE model on his estimate. In this study his estimate of the demand elasticity for gasoline has a standard error of approximately thirty percent (or 0.3). The standard error of an estimate is a good indicator of the volatility of that parameter over making it well-suited to serve as a measure of gasoline demand variation in our Monte Carlo study.

⁶ Finding a volatility measure here is a challenging task since there are no readily available estimates of production costs or cost variation in Mexico. Economic theory, however, suggests that, in competitive equilibrium, market prices should serve as a good proxy for production costs. Furthermore, economic agents’ expectations regarding market conditions and market volatility are embodied in futures prices and the variation in those prices. Using data from the US Energy Department (US Department of Energy, 2018) then, we examine the variance of the US ethanol futures market over the next 25–35 years, and obtain a standard deviation of 0.2. The estimate obtained her should not, of course, be considered exact. It does, however, fulfil our more modest goal of obtaining a variation parameter which has the correct order of magnitude, making it suitable for carrying out a Monte Carlo simulation exercise such as this.

⁷ We also ran the model assuming uniform distributions to test the robustness of the results. This, however, did not significantly change the nature of our output and in the interest of space the uniform distribution results are not included here. These results, however, may be obtained from the authors upon request.

⁸ Since consumer demand equations in our CGE model are derived from their underlying (nested) CES expenditure functions. The variation of demand elasticity for goods involves the variation of the elasticity of substitution between consumer goods within the utility functions of our four classes of consuming agents.

⁹ We show the results from the normal distribution only. Results on the uniform distribution are not qualitatively different, therefore are not shown in the figures. However, they are available upon request to the authors.

¹⁰ In their analyses of Ethanol production (Cheali et al. (2016) Cheali et al. (2014)) conduct a thorough engineering-based cost analysis and consider a number of different biofuel production techniques. They deal with uncertainty and employ Monte Carlo techniques to determine those production processes which have the best chances of market success. Their cost estimates vary by + – 30%. This result is encouraging since variation of this order of magnitude is well bracketed within the two tails of the randomly generated production cost distribution which we have constructed here.

¹¹ It should be noted that this exact 50/50 breakdown is somewhat dependent on our decision to use the minimum subsidy necessary to induce activity in the slack (i.e. ethanol) sector. Had we employed a higher subsidy then ethanol would have been a slack activity in less than 50% of our Monte Carlo runs. The nature of our results, however, would have remained unchanged.

¹² Again, results from uniform distributions are not qualitatively different. Therefore, they are not shown.