Parametrical regulation of multi-criteria economic policy based on nonautonomous computable general equilibrium models Abdykappar A. Ashimov, Yuriy V. Borovskiy, Bahyt T. Sultanov, Rakhman A. Alshanov, Nikolay Yu. Borovskiy, Bakytzhan A. Aisakova Kazakh National Technical University, Almaty, Republic of Kazakhstan Abstract. In paper the problem of a choice of the effective concerted state policy in the sphere of economic growth with the help of multi-criteria optimization on the basis of two nonautonomous computable general equilibrium models (CGE models) with the common economic optimization tools is considered. Conditions of consistent optimization on the base of pair of models and development of the parametrical regulation theory elements on a class of discrete non-autonomous controllable dynamic systems are given. Efficiency of application of one method of joint parametrical identification of two investigated non-autonomous models (CGE model of sectors of economy and CGE model with knowledge sector) is shown. The parametrical identification of the considered pair of investigated macroeconomic mathematical models consists in finding estimations of unknown parameters (values of exogenous functions) in the given limited areas for which the minimum value of some target function (criterion) is reached. The estimation of Pareto set for considered two-criteria problem and the solution (on the base of Pareto set) of an optimization problem with one utility function are found. Key words: Computable general equilibrium model, Parametrical identification, Multicriteria optimization. 1. Elements of the parametrical regulation theory on the basis of discrete nonautonomous dynamical systems We consider the following discrete controllable system: π₯(π‘ + 1) = π(π₯(π‘), π’(π‘), π(π‘)), π‘ = 0, 1, … π − 1; π₯(0) = π₯0 . (1) Here t – time, taking non-negative integer values; π₯ = π₯(π‘) – vector-function of the systems’ state; π’ = π’(π‘) – control; vector-function of a discrete argument; π = π(π‘) – known uncontrollable vector-function of a discrete argument (element of the Euclidian space of the corresponding dimension); π₯0 – known initial state of the system; π – known vector-function of its arguments. The method of choosing optimal values of economic tools is associated with the model of maximization of the optimality criterion max ∑ππ‘=1 πΉ[π‘, π₯(π‘)], π’ 1 (2) where πΉ – known function and with the following constraints. Phase constraints imposed on the solution of the system (1) are of the following type: π₯(π‘) ∈ π(π‘), (π₯(π‘)) ∈ ππ (π‘), π‘ = 1, … , π, (3) where π – given vector-function; π(π‘), ππ (π‘) – given sets. Constrains imposed on control are as follows: π’(π‘) ∈ π(π‘), π(π’(π‘)) ∈ ππ (π‘), π(π’(0), π’(1), … , π’(π − 1)) ∈ ππ , π‘ = 0, … , π − 1, (4) where π, π – given vector-functions; π(π‘), ππ (π‘), ππ – given sets. On the basis of the relations (1) – (4) we obtain the following variational problem, so called the variational calculus problem on the synthesis of optimal laws of parametrical regulation. The analogous problem and theorems, corresponding to the described below theorems 1 and 2, for the case of autonomous dynamical systems were considered in (Ashimov et al, 2012). Problem 1. Given the known function π find control π’, satisfying the condition (4), so that corresponding to it solution of the dynamical system (1) satisfies the condition (3) and provides the maximum for the functional (2). Let ππ – set of allowed "state-control" pairs of the considered system under the given known function π, i.e. the set of such “vector-function” pairs (π₯, π’), which satisfy the relations (1), (3), (4). We introduce notations: π = βππ‘=1 π(π‘), π = βπ−1 π‘=0 π(π‘). The following two theorems are true, their proofs are based on application of continuous functions’ features, and particularly, on application the features of the functions continuous on the compact. Theorem 1. Let for the known function π the set ππ be nonempty; sets π(π‘) and π(π‘ − 1) be compact, and sets ππ (π‘), ππ (π‘ − 1) be closed for all π‘ = 1, … , π; the function πis continuous in the first two arguments on the set π × π, and the function πΉ is continuous in the second argument on the set π. Let the mappings π, π, π be continuous on the sets π, π, ∏π−1 π‘=0 π(π‘) respectively; the set ππ is closed. Then the problem 1 has a solution. Theorem 2. Let the conditions of the theorem 1 be met for any functions π ∈ π΄ (where π΄ – a open set in the Euclidian space); the function π is continuous in the third argument in A and satisfies the Lipschitz condition in the first argument in X uniformly, in the second and third arguments in π × π΄. Then the optimal value of the criterion of the problem 1 continuously depends on uncontrollable function π from π΄. 2. Choice and the joint parametrical identification of the two CGE models The following requirements are the basis for selecting two or more mathematical models for solving the problems of consistent optimization. 1. Acceptable accuracy of the model description of a single subject area, in the framework of which the problem of consistent optimization is solved with the criteria (πΎ1 , πΎ2 , …), corresponding to the selected targets of economic policy, as well as the possibility of obtaining these criteria. 2 2. Compliance with the requirements of the formulation and solution of multi-criteria optimization problem based on the considered models. 3. Effective realization of solutions to the consistent optimization problem. Within the first requirement we consider the possibility of choosing two CGE models for the consistent optimization in the sphere of economic growth. This possibility is evaluated by solving the problem of joint parametrical identification of the following models (Makarov et al, 2007). Model 1. CGE model of economic sectors is presented by 16 economic producing agents (sectors), as well as by non-productive sectors: total consumer, government and banking sector. This model contains 698 equations, with the help of which the values of its 698 endogenous variables are calculated. It also contains 2045 estimated exogenous parameters (values of exogenous functions). Model 2. CGE model with knowledge sector is presented by the following six economic agents (sectors): sector of science and education (knowledge), innovative sector, representing the set of innovative-active enterprises and organizations; other sectors of economy; aggregate consumer, uniting households; government; banking sector. Here the first three economic sectors are producing agents. The model with knowledge sector contains 110 endogenous equations, with the help of which the values of its 110 endogenous variables are calculated. This model also contains 86 estimated exogenous functions. Out of them 9 exogenous functions are common for the two considered models. The problem of parametrical identification of the considered pair of researched macroeconomic models comes to finding the estimates of unknown parameters (values of exogenous functions) in the restricted areas specified for each model, under which we reach a minimum value of the objective function πΎπΌ , characterizing: - deviations of the calculated values of endogenous variables of each model from the corresponding observed values (known statistics); - differences between the corresponding values of l = 18 endogenous variables of the models 1 and 2, which have the same meaning; and with the additional condition of coincidence of the corresponding values of estimated parameters common to both models. As the result of numerical solution of parameter identification based on the statistics of the Republic of Kazakhstan for 2000 – 2008, applying two criteria (Ashimov et al, 2012) and the Nelder-Mead algorithm (Nelder and Mead, 1965) the value πΎπΌ = 0.026 was obtained. This number is equal to the root-mean-square value of the relative deviations and discrepancies, presented above. 3. Consistent optimization on the basis of the two CGE models In the framework of the requirements 2 and 3 presented above (Section 2) the paper considers multi-purpose choice of optimal economic policy in the sphere of economic growth and in the sphere of innovation with the two-dimensional criterion πΎ = (πΎ1 , πΎ2 ). The components of the πΎ criterion respectively describe the mean value of real GDP (πΎ1 ) of the first model and the mean value of real gross value added of the innovative sector (πΎ2 ) of the second model. The considered problem of two-criterion optimization has the following form (5)-(7). 3 max πΎ(π). π (5) Here π = {πππ (π‘): π = 1, … ,16; π = 1, 2, 3; π‘ = 2011, … , 2015 } – the set of common regulated parameters; πππ (π‘) – additional investments in year t to the j-th sector of the first model within the framework of subsidizing the k-th sector of the second model (or the same, additional investments to the k-th sector of the second model in year t within the framework of subsidizing the the j-th sector of the first model). The constraints of the type (4) imposed on the controlled parameters and the conditions of its realization on the basis of the two models are as follows: πππ (π‘) ≥ 0; ∑π‘,π,π πππ (π‘) ≤ 11.5 × 1012 ; ππ (π‘) = ∑π πππ (π‘); ππ (π‘) = ∑π πππ (π‘). (6) Here 11.5 × 1012 – total volume of investments in tenge for the period 2011 – 2015 (tenge – national currency of Kazakhstan); ππ (π‘) – additional investments in year t to j-th producing sector of the first model; ππ (π‘) – additional investments in year t to k-th producing sector of the second model. Obviously, the stated additional investments satisfy equality ∑π ππ (π‘) = ∑π ππ (π‘) = ∑π,π πππ (π‘), which guarantees equal amounts of additional investment, consistently invested in the economy for each t year of regulation in the framework of each of the two CGE models. Constraints of the form (3) imposed on the vectors of endogenous variables π¦1 (π‘) and π¦2 (π‘) of the first and the second model respectively, as well as imposed on l-th having the same meaning the coordinates of these vectors, have the following form: π¦1 (π‘) ∈ π1 (π‘), π¦2 (π‘) ∈ π2 (π‘), |π¦1π (π‘) − π¦2π (π‘)| ≤ π π . (7) Here π1 (π‘) ΠΈ π2 (π‘) – given sets; π π – given small amounts, π = 1, … , 18, π‘ = 2011, … , 2015. For the formulated two-criterion problem applying the Nelder-Mead algorithm the estimate Π of the Pareto set (see Figure 1) was obtained as the set of points sequentially connected by a continuous line on the plane (πΎ1 , πΎ2 ). Units of criteria measurements on the figure: πΎ1 – 1012 tenge, πΎ2 – 1010 tenge in prices of 2000. Existence of solutions to auxiliary (for estimating Π) single-criterion optimization problems (max πΎπ (π), π = 1,2), relating to the type of problem 1 π with corresponding constraints imposed on endogenous variables and regulated parameters, as well as continuous dependence of corresponding optimal criterion values on uncontrolled functions are justified with the help of Theorem 1 and 2. Here the selected pair of consistently controlled CGE models is used as discrete controlled system (1). The solution to optimization problem of linear utility function of the form π(πΎ1 , πΎ2 ) = 2 ∑π=1 ππ πΎπ /πΎπ0 for the case π1 = π2 = 0.5 (equal importance of both criteria for the decision maker) and for selected basic values of the criteria πΎπ0 is presented in Figure 1 with the help of the point π΄(6.27, 4.30) ∈ Π. The sloping straight line passing the point π΄ is a equipotential line of the utility function π for the case corresponding to the maximum value of this function on the set Π. 4 Figure 1. Estimation of the Pareto set References Ashimov, A.A., Sultanov, B.T., Adilov, Zh.M., Borovskiy, Yu.V., Novikov, D.A., Nizhegorodtsev, R.M. and As.A. Ashimov (2012); Macroeconomic analysis and economic policy based on parametric control; New York: Springer (pp. 265). Makarov, V.L., Bakhtizin, A.R. and S.S. Sulakshin (2007); The use of computable models in public administration; Moscow: Scientific Expert, (pp. 304, in Russian). Nelder, J.A. and R. Mead (1965); A simplex method for function minimization; The Computer Journal, No. 7, (pp. 308–313). 5