Environmental and Resource Economics Methods: Dynamics and Inter-temporal decisions KLAUS EISENACK - RESOURCE ECONOMICS GROUP HUMBOLDT-UNIVERSITÄT ZU BERLIN WWW.RESOURCE-ECONOMICS.HU-BERLIN.DE Mathematical tools to deal with time, stocks and flows Klaus Eisenack 2 Stocks and flows, mathematically (recalled from intro) • Flow πΉπΉπ‘π‘ in time π‘π‘ • Stock πππ‘π‘ with endogenous change rate ππ ππ , described as a mathematical function • Dynamics described by difference equations: πππ‘π‘+1 = πππ‘π‘ + ππ πππ‘π‘ + πΉπΉπ‘π‘ f(S) “fish stocknext year = fish stockthis year + fish growththis year – catchthis year ” “atmospheric carbonnext year = atmospheric carbonthis year – decaythis year + emissionsthis year ” Klaus Eisenack (example) S 3 A second way to describe change: Differential equations (ordinary differential equations, ODE) • Comparison to difference equations xt +1 = xt + f ( xt , ut , t ) − While time evolves step-by-step (e.g. years, months, days) in difference equations, time evolved continuously for differential equations • Describe processes over time; applications in many scientific disciplines, e.g. − Physics: mechanics, movement of bodies, radioactive decay − Chemistry: course of chemical reactions − Biology/ecology: population growth, ecosystem dynamics − Demography/epidemiology: spreading of diseases, demographic change − Economics: growth theory, resource economics • Basic idea: the change rate of a system state over time dx/dt (derivative with respect to time, „velocity “) at a given time t depends on exogenous parameters u und the state x itself. Dynamics starts at time t0 with in state x0: xο¦ = dx dt = f ( x, u , t ), x( t 0 ) = x0 • For example: non-renewable resource stock x, extraction share u Klaus Eisenack xο¦ = −u ⋅ x, x(0) = 1000 4 Differential equations (ctd.) • Solving a differential equation: xο¦ = f ( x, u , t ), x(0) = x0 • Numerical approximation: “Start from the initial position x(0) and the initial strategy u(0). Determine the speed f(x,u,0). Continue with this speed for a marginal instance ε. You then arrive at the position x(ε)= x(0)+ ε ⋅ f(x,u,0). Now continue with the strategy u(ε), and so on and so forth.” • Algebraically: “Find a function x(t) in time with the following property: If x(t) is plugged into f (together with u(t)), this yields exactly the derivative of x(t).” • Note: explicit solutions are usually difficult or even impossible to find, but important properties can frequently be derived There are various sources that help you to learn ODEs. In this class, you only need the very basics. For example: https://www.youtube.com/watch?v=f8xtjVGnnXo https://www.khanacademy.org/math/ap-calculus-bc/bc-differential-equations-new/bc7-5/v/eulers-method Klaus Eisenack 5 Differential equations: example Exponential growth • A physical capital stock x(t) at time t produces µ·x(t) units of goods that can be invested to increase the capital stock. The initial capital stock is x(0)=x0. − How does the capital stock develop over time if its product is completely reinvested? • Differential equation: xο¦ = µ x • The solution is a function x(t) with the following properties: x(0) = x0 , and for all t ≥ 0 : dx ( t ) dt = µ x(t ) • Exercise: show that the exponential function is the solution: x (t ) = x0e µt Klaus Eisenack 6 Intertemporal decisions: first thoughts Klaus Eisenack 7 Thought experiment: cake eating problem • How to invest in and make use of a capital stock over time? • Cake eating problem: simplest version − Robinson Crusoe has saved 1000 pieces of zwieback from the ship, and stays on the island for two years: how much should he eat in the first, and how much in the second year? − Think of extensions: 40 years instead of two; uncertain time on the island; … • One way to solve the problem: optimal plan (utilitarian) − Cakes in year one (R1), and in year two (R2), maximize total utility W(R1,R2) subject to R1+R2=1000 ππππ ππππ − Lagrangean yields πππ π =πππ π 1 2 ο Equalization of marginal utility over both periods (vgl. Ströbele, 1987, Rohstoffökonomik, Verlag Franz Vahlen, p. 15-18) Klaus Eisenack 8 Intertemporal efficiency and optimality • Recall standard Pareto efficiency… • Dynamic Pareto efficiency − You cannot extract more of a resource at one point in time without sacrificing extraction at another time • Static optimality / allocative efficiency − A welfare function is maximized (at each time) − Marginal net benefits of all goods and services equalized (at each time) • Intertemporal optimality / dynamic efficiency − A welfare function that aggregated benefits over time is maximized − Marginal net benefit of a resource equalized over time Perman et al. Ch. 11.1, Ch. 14.5.2 Klaus Eisenack 9 Discounting: how to aggregate benefits over time Pe r m a n e t a l . C h . 3 . 5 Complementary reading: R . P. M c A fe e , T. L e w i s , D. D a l e ( 2 0 1 6 ) I n t ro d u c t i o n t o E c o n o m i c A n a l y s i s , M i c ro s o f t / D u ke U n i v e rs i t y / M u h l e n b e rg C o l l e g e , h tt p s : / / w w w. ke l l o g g . n o r t h w e st e r n . e d u / fa c u l t y / d a l e / i e av 2 1 . p d f [ 2 2 . 4 . 2 0 ] , C h . 1 1 . 1 C o r e Te a m ( 2 0 1 7 ) T h e Ec o n o my, O n l i n e E b o o k , h tt p : / / w w w. c o re - e c o n . o r g / , U n i t s 10.3, 10.4. Klaus Eisenack 10 Comparing present and future utility: Discounting and net present value • Usually: present utility of a given quantity larger than future (expected) utility of the same quantity, but valued today W = U ( R0 ) + 1+1ρ U ( R1 ) + ( ) U ( R ) + ... + + ( ) U (R ) + − Time preference / patience − Uncertainty about future conditions (risks, death, doomsday, technological progress) − Economic growth − Current value(s): U(R0), U(R1), …, U(RT); (Net) present value: W • To consider time preference for present consumption: derive net present value W from stream of current value utility U(·), using a discount rate ρ 1. (discrete time) 2. (continuous time) • For private investments: consider a stream of profits π(t) and a market interest rate i 1. (discrete time) 2. (continuous time) T W =∑ t =0 T 1 2 1+ ρ 2 1 T 1+ ρ T ( ) U (R ) 1 t 1+ ρ t W = ∫ U ( R(t ))e − ρt dt W = ∑ (1+1 i ) tπ (t ) t =0 T 0 T W = ∫ π (t ) e −it dt 0 Discount admits to compare intertemporal decisions Example • Suppose a firm can choose between two investment options, which both cost the same amount • Both investments promise a certain stream of profits over 4 years (see table) • If profits are put to a bank account, the firm earns an interest rate of 10% per year. ο Which option is the best for the firm? • Exercise: determine and compare the present value of both options year profit (option 1) (See supplied excel sheet for a way to solve the task) 0 100 80 1 100 100 2 100 120 3 100 140 Klaus Eisenack profit (option 2) 12 Dynamic optimization: determining optimal intertemporal decisions about dynamic systems Perman et al. Ch 14.5 and Appendix 14.1 Compementary reading: Ströbele (1987) Rohstoffökonomik, Vahlen, p. 15-21, 175-177 Dynamic optimization (or: dynamic control) • Problem: Find a control path in time π’π’(οΏ½), so that ππ max ππ π’π’(οΏ½) β ∫0 πΌπΌ π₯π₯ π‘π‘ , π’π’(π‘π‘) ππ −πππ‘π‘ ππππ π’π’(οΏ½) subject to π₯π₯Μ = ππ π₯π₯ π‘π‘ , π’π’(π‘π‘) , π₯π₯ 0 = π₯π₯ 0 for all π‘π‘ ∈ 0, ππ : π’π’ π‘π‘ ∈ ππ (and further technical assumptions) • Cake Eating Problem? • Many applications, e.g. π₯π₯(π‘π‘): state of system at time π‘π‘ π’π’(π‘π‘): strategy/control chosen at π‘π‘ from admissible set ππ πΌπΌ(π₯π₯, π’π’): current value at state π₯π₯ if strategy π’π’ is chosen ππ: discount rate ππ: present value if control path π’π’(οΏ½) is chosen − Engineering: mechanics with minimal energy costs − Physics: movement of bodies in mechanical systems − Military: rockets reaching their target in minimal time − Economics: optimal growth, optimal resource exploitation, dynamic game theory Klaus Eisenack 14 Cake eating problem, revisited • Robinson Crusoe can dispose of 1000 packs of zwieback • He assumes an annual 10% probability to be rescued • How much packs of zwieback shall he consume per year? − Further assumption: isoelastic current value of zwieback • Formulation as dynamic optimization problem ∞ max W := ∫ U (C (t ) ) e − µt dt , C ( ⋅) 0 s.t. Sο¦ = −C , S (0) = 1000, C ≥ 0, with U = 1−1a C 1− a , µ = 10% Klaus Eisenack 15 Dynamic optimization: „cooking recipe“ • Introduce a dynamic shadow price λ(t) and Hamiltonian H (cf. Lagrange) • (1) For every point in time: maximize H with respect to u(t) • (2,3) Differential equations for x and λ • (4) transversality condition (These conditions are necessary but not always sufficient) Ideas: • • • Reduce the intertemporal problem to problems (1) that can be solved for each time t independently The role of intertemporal trade-offs it taken over by λ The shadow price λ „integrates back“ all future value of the state variable (see 3, 4). Klaus Eisenack H ( x, u , λ , t ) = I ( x, u , t ) + λ ⋅ f ( x, u , t ) (1) for all t ∈ [0, T ] : max H (x(t ), u (t ), λ (t ), t ) u (t ) (in the simple case : dudH( t ) = 0) (2) xο¦ = dH dλ = f ( x, u , t ) (3) − λο¦ + ρλ = dH dx (4) lim λxe − ρt = 0 t →T 16 Let us solve the cake eating problem (solution) • “Cooking recipe” H = U (C ) − λC , dH ο¦ = −C , S (0) = S , S = 0 , 0 dC − µt µλ − λο¦ = dH λ t S t = 0. , lim ( ) ( ) e dS t →∞ • Leads to the solution: λ (t ) = λ0 e µt , lim S (t ) = 0, t →∞ µ C (t ) = a S 0 e − µa t . • All cake is eaten, with decreasing rate • Exponentially increasing shadow price = constant marginal value of resource (in present value) • Concrete solution with µ=0.10, a=½: C(t)=200 e-0.2t Klaus Eisenack 17 Klaus Eisenack 18 Core reading and guiding questions • Perman et al. Ch 3.5, 11.1, 14.5 and Appendix 14.1 • Compare difference and differential equations: what pros and cons do you see for the two approaches? • Show that π₯π₯ π‘π‘ = 2 οΏ½ ππ −3οΏ½π‘π‘ + 1 is a solution of the ODE π₯π₯Μ = −3π₯π₯ + 3. • Dynamic Pareto efficiency and intertemporal optimality: does one of these criteria imply the other one? Which criterion is more strict? • Solve the exercise to compare two decision alternatives by determining their net present value • Dynamic optimization: − Solve the supplementary training tasks on Moodle − Describe another problem of your choice as a dynamic optimization problem (you don’t need to solve it) − Compare dynamic optimization with the Lagrange approach: what is similar, what different? Klaus Eisenack 19