Estimation of food demand in Turkey using quadratic almost ideal demand system

Demand estimation is crucial for people, companies, and governments. It is important not only for predicting the future quantities, but is also important to understand the consumer behaviour. Because, understanding the consumer behaviour helps to decision-making process. This thesis estimates the food demand in Turkey for the years between 2008 and 2016. Because, drastic increases in inflation rates in recent years is mostly attributed to the food prices. Besides, demand for food is relatively more inelastic compared to the other commodity groups. So, food demand in Turkey is an important subject for the welfare of the Turkish people. One can make inferences on the consumers' well-being by observing the revealed preferences which is a subject of the Microeconomic Theory. By understanding the Neoclassical Demand Theory, one can better understand the consumers' behaviours (and preferences). Therefore, I examine the Neoclassical Demand Theory before estimating the food demand in Turkey. I use Almost Ideal Demand System to estimate food demand, because it satisfies many desirable assumptions of the Neoclassical Demand Theory. However, I make estimations using Quadratic AIDS instead of AIDS. Because QUAIDS assumes that consumers have quadratic Engel Curve which is more appropriate for the demand systems. On the other hand, AIDS and QUAIDS are nonlinear models; but in order to make estimations faster, it is converted into a linear model called Linear Approximated AIDS. However, LA/AIDS has some econometric problems like endogeneity; therefore I estimate QUAIDS nonlinearly. One of the distinctive feature of this study is that I estimate food demand after aggregation of household surveys. I convert surveys into time series data. By doing so, I notice unusual behaviour of the consumers during the sacrifice feasts; I estimate the model faster; and I do not need to apply censored demand techniques to reduce heterogeneity. However, the number of observations is reduced to 108 from (almost) 100.000 by aggregation. Thus, I make robustness checks at some points. The results carry conviction and give important intuitions about the current debates like ``bread pricing'' and ``meat imports''. ``Meat and Fish'' and ``Sugar, jam, honey, chocolate, and confectionary'' commodity groups have high income elasticity. Thus, welfare effects can differ for the low and high income people. The difference between compensated and uncompensated price elasticity for ``Bread and Cereals'', ``Fruit and Vegetables'', and ``Meat and Fish'' commodity groups are very high. This means, consumers cannot substitute these commodity groups. Thus, price increases for these commodity groups yield high welfare loss. In order to make better welfare analysis, one should estimate the food supply and supply elasticities of the commodity groups. And further methodological researches can be conducted on the machine learning algorithms. ML algorithms offer better performance of prediction accuracy and precision. However, there is a huge literature gap on interpretability of the ML algorithms. ML can be used for demand systems as well.