The history of macroeconometric model-building is comprehensively documented in Fair (1984, 1994), Bodkin et al. (1991), Hendry and Mizon (2000), Favero (2001), Pagan (2003a, b), Bårdsen et al. (2004, 2005), Valadkhani (2004), Hendry and Muellbauer (2018), Jelić and Ravnik (2021) inter alia. Also, history and macroeconometric modeling activities over the world and their classification are documented in Welfe (2013).

This section reviews only general equilibrium macroeconomic models that have been built for the Saudi Arabian economy. In other words, we do not review either partial equilibrium models built for Saudi Arabia (e.g., see Mohaddes et al. 2020) or general equilibrium models built for other resource-rich economies. The former ones are not in line with our objectives, and the latter ones are out of the scope of this book and have been reviewed by Welfe (2013) and Hasanov and Joutz (2013) to some extent, among others. Our review here is limited to models that are publicly available or available to us.Footnote 1 Table 2.1 documents these models.

Table 2.1 Macroeconomic models for Saudi Arabia
Full size table

As the strengths and weaknesses of each model are documented in Table 2.1, we do not discuss them again here. However, it is worth mentioning that their strengths and weaknesses are also determined by their type that they belong to among other factors. In general, structural, that is theory-guided models, such as computable general equilibrium (CGE) and dynamic stochastic general equilibrium (DSGE) models have the main strengths of being strongly consistent with textbook economic theory, useful for long-term projections and analyzing the effects of changes in policy variables. The studies listed below discuss that these models have the following main weaknesses: using micro-foundations strictly as theoretical foundations and not allowing data ‘to speak freely’; they do not incorporate information about behavioral economics and information economics; they are calibrated to capture only equilibrium positions with none to limited information about short-run dynamics, and they do not provide information about the errors that they make in their representations and simulations; they rely on many assumptions, restrictions, parametrizations that are not always true in reality (see Romer 2016; Stiglitz 2018; Blanchard 2017, 2018; Hendry and Muellbauer 2018; Wren-Lewis 2018; Fair 2019; Colander et al. 2008; Colander 2006; Hara et al. 2009; Pagan 2003a; Gürkaynak and Tille 2017; Crump et al. 2021; Wickens 1995 inter alia). Additionally, Giacomini (2015), Gürkaynak et al. (2013), among others, show that DSGEs, pure structural models produce very poor forecasting performance compared to econometric models in the empirical analyses. Moreover, Wickens (1995), Pesaran and Smith (2011), Blanchard (2017), inter alia, discuss that for DSGE models to survive in the future, they should account for data and hence switch from calibration of the deterministic relationships to estimation of the stochastic specifications, they should estimate well-specified long-run relationships rather than trend approximations and consider more dynamic short-run specifications to possibly account for habits, expectational errors, learning, and the costs and frictions of search and matching, and they should relax the assumptions made, such as optimal behavior, homogenous agent, symmetric information about market conditions, etc. Furthermore, Nikas et al. (2019, p. 37–38) discuss that standard structural models assume that markets clear in the short-run and, hence, they ignore disequilibrium and short-run relationships. For example, they usually assume that there is no unemployment in their representation of an economy. This obviously is not a relevant assumption even in the long-run and, hence, leads to drawbacks in their performances. Most likely due to the above-mentioned issues, the government agencies such as central banks recently prefer hybrid type macroeconometric models, which are built using equilibrium correction equations, in their policy analyses, forecasting, and projections. Because hybrid macroeconometric models perform better than purely theory-based models (e.g., CGE, DSGE, optimal growth models) and purely data-based models (e.g., unrestricted vector autoregression models) since they are the combination of theory-guided and data-driven approaches as the literature discusses (see discussions in Ballantyne et al. 2020, Cusbert and Kendall 2018, Hendry 2018, Hendry and Muellbauer 2018, Bulligan et al. 2017; Jelić and Ravnik 2021; Giacomini 2015; Pagan 2019; Gervais and Gosselin 2014). Moreover, the behavioral representations of economic agents in the macroeconometric models are based on their historical evolution, whereas in the theory-guided models, they are usually based on the optimization of a representative agent, imposed parameters, and calibration using data from a single year or an average of years (e.g., see Lutz 2011; Lehr et al. 2012).

Jelić and Ravnik (2021) and Pagan (2019), among other studies, discuss four generations of macroeconometric models that are coexisted for the last more than 80 years and recent hybrid models incorporate the insights derived from the third- and fourth-generation models into the second-generation models. The main strengths of the hybrid types of macroeconometric models (MEMs) over the other types of macroeconomic models are that they have theoretical coherence to represent long-run equilibrium relationships (like CGE and DSGE models and unlike VAR models). They also possess empirical coherence, i.e., they allow the data ‘to speak freely’ (unlike CGE and DSGE models and like VAR models) to represent short-run dynamics and disequilibrium. In other words, they bring together ‘theory-guided’ and ‘data-driven’ approaches (e.g., see Hendry 2018). They can represent the behavioral aspects of economic relationships based on the statistical time series properties of national data. Other advantages of MEMs are that they can be modified or customized to accommodate different policy questions and various simulations can be done in one model simultaneously, making them user-friendly for policy analyses. Their main weaknesses are, as mentioned in the Table 2.1, being data-dependent, data updates and revision issues require a reconsideration of all the behavioral equations, require a large team for data and model maintenance and update. For detailed strengths and weaknesses of different kinds of models, interested readers can refer to the above-listed references as well as Ackerman (2002), Pagan (2003b), Hoover et al. (2008), Herbst et al. (2012), Arora (2013), Hurtado (2014), and Oxford Economics (2022).

KGEMM is a hybrid model, i.e., it combines an economic theory-guided modeling approach with empirical data-driven evidence.Footnote 2 This is performed through statistical estimations and testing, not by imposing theory on the model. Practically, it attempts to adjust for econometric weaknesses in earlier models built for Saudi Arabia. KGEMM also incorporates detailed demand-side representations and CO2 emissions of the main energy products by customer type. In this regard, KGEMM is a type of E3ME model (Energy-Environment-Economy Macro-Econometric model, see Econometrics, Cambridge 2019; Nikas et al. 2019; Gramkow and Anger-Kraavi 2019; Lee et al. 2018; Dagoumas and Barker 2010, inter alia). And it is similar to SEEEM (Sectoral Energy-Economic Econometric Model, see Blazejczak et al. 2014a, b) and PANTA RHEI (see Lutz et al. 2014a, b; Flaute et al. 2017; Lehr and Lutz 2016; Lehr et al. 2012; Lutz 2011), which both cover energy-economic-environmental representations.