Dynamic Adaptive Policymaking: An Approach to Planning Under Deep - PowerPoint PPT Presentation

Dynamic Adaptive Policymaking: An Approach to Planning Under Deep Uncertainty Prof. Dr. Warren E. Walker New Zealand Climate Change Research Institute Victoria University of Wellington 10 February 2015

The Problem: How to make policy in a deeply uncertain world • “There are things that we know that we know; there are things that we know that we don’t know; there are things that we don’t know that we don’t know.” – Donald Rumsfeld, 2/12/2002 • How do we make policy taking into account the things that we know that we don’t know, and safeguard against/prepare for the things that we don’t know that we don’t know? 2

Outline • The traditional (‘ Predict-and- Act’) approach to policymaking • What is ‘deep uncertainty’? • Why the traditional approach does not work under deep uncertainty • The inverted ‘Monitor -and- Adapt’ approach to policymaking (with Dynamic Adaptive Policymaking (DAP) as an example) 3

A Framework for Traditional Policy Analysis 4

Actors Involved in a Policy Analysis Study Stakeholders Problem owner(s) (policy arena actors) (decisionmaker(s)) Policy analysts (decision advisors) Methodologists (toolmakers) Specialists (domain experts) 5 2/16/2015 09:53

Summarizing the Traditional Approach to Policy Analysis* * Warren E. Walker, “Policy Analysis: A Systematic Approach to Supporting Policymaking in the Public Sector”, Journal of Multicriteria Decision Analysis , Volume 9, No. 1-3 (2000), pp. 11-27. 6

Among the Assumptions • The future context(s) • The system boundary • The system model (elements; relationships among elements) • The relative importance of the outcomes 7

Large Losses Possible if Take Action Based on Wrong Predictions 8

Large Needless Expenditures Possible if Take Action Based on Wrong Predictions 2,200 1975 Conservation scenarios 2000 Actual 2,000 1,800 1,600 1990 Gross 1,400 national Historical product, 1,200 trend 1980 continued USA 1,000 1977 (billions 1973 800 of 1958 1890 1970 600 1900 dollars) 1930 1960 400 1950 1940 200 1920 1929 1910 0 0 20 40 60 80 100 120 140 160 180 Energy use (10 15 Btu per year) 9

Lessons About Predictions • “Predictions are very difficult, especially about the future.” -- Yogi Berra • “In practice, decisionmaking is always in the face of uncertainty… Waiting for an impossible absolute truth means never doing anything.” -- D. Bradbury, letter to The Times , 6 January 2005 • Implementing policies that assume that the future will be a continuation of the past is like driving a car while looking only into the rear view mirror. • You cannot choose a scenario. The probability that any specific scenario will be correct is zero. 10

The Locations of Uncertainty in a Policy Analysis Study (X) Context (R) System model (O) System outcomes (W) Weights on the outcomes 11

Uncertainties are Found in the X,R,O,W Locations • Uncertainties about the future (demographic, economic, social developments, etc.) • Uncertainties about the system (since (some of) the key relationships determining system performance are insufficiently known) • Uncertainties about the policy outcomes • Uncertainties about the valuation of the outcomes by different stakeholders 12

Model-Based Decision Support Must Deal With Several Levels of Uncertainty at Each of the Locations LEVEL Level 1 Level 2 Level 3 Level 4 Context A clear enough Alternate futures A multiplicity of Unknown future (X) future (with probabilities) plausible futures Complete Certainty Total ignorance LOCATION System A single A single Several system Unknown system Model (R) (deterministic) (stochastic) system models, with model; know we don’t know system model model different structures System A point A confidence A known range Unknown Outcomes estimate for interval for each of outcomes outcomes; know we don’t know (O) each outcome outcome Weights A single set of Several sets of A known range Unknown weights; know we don’t on weights weights, with a of weights outcomes probability know (W) attached to each set 13

Today I Will Focus on Uncertainty About the Future Context LEVEL Level 1 Level 2 Level 3 Level 4 Context A clear enough Alternate futures A multiplicity of Unknown future (X) future (with probabilities) plausible futures Complete Certainty Total ignorance LOCATION System A single A single Several system Unknown system Model (R) (deterministic) (stochastic) system models, with model; know we don’t know system model model different structures System A point A confidence A known range Unknown Outcomes estimate for interval for each of outcomes outcomes; know we don’t know (O) each outcome outcome Weights A single set of Several sets of A known range Unknown weights; know we don’t on weights weights, with a of weights outcomes probability know (W) attached to each set 14

Most Approaches for Dealing with Uncertainty about the Future are Problematic • Ignore uncertainty • Assume the future is knowable (‘predict -and- act’  “optimal” policy) (Level 1) • Assume the future will (probabilistically) look like the past (driving while looking only through rear-view mirror) (  ‘trend - based’ policy)(Level 2) • Look for a policy that will do well in a few scenarios (  ‘static robust’ policy) (Level 3) • What if the experts do not know and /or stakeholders can not agree on what the future might bring (Level 4) = “ deep uncertainty ” 15

Definition of Deep Uncertainty* A situation in which the relevant actors do not know, or cannot agree upon: – how the system works – how likely or plausible various future states are – how to value the various outcomes of interest * Walker , Warren E., Robert J. Lempert, and Jan H. Kwakkel (2013). “Deep Uncertainty”, entry (pp. 395 -402) in Saul I. Gass and Michael C. Fu (eds.), Encyclopedia of Operations Research and Management Science., 3 rd Edition , New York: Springer. 16

Four Approaches for Dealing with Deep (Level 4) Uncertainty • Resistance: plan for the worst possible case or future situation – Likely to be very costly • Resilience: whatever happens in the future, make sure that the system can recover quickly – Accepts short-term pain; focuses on recovery • Static robustness: aim at reducing vulnerability in the largest possible range of conditions – May be difficult to change when conditions change • Planned adaptation: plan to change over time, in case conditions change 17

An Approach for Planned Adaptation: Dynamic Adaptive Policies (DAP) 18

DAP Extends the Analysis

Features of an Adaptive Policy • Tests assumptions (in the real world) • Hedges against negative outcomes and other uncertain events • Steers toward positive outcomes • Adapts to changing situations 20

The DAP Approach (1) • Determine a set of goals • Identify a promising basic policy and conditions for its success • Identify vulnerabilities of the policy (how it could fail) and ways of protecting it • Monitor progress toward the goals • Adapt the policy as exogenous and endogenous conditions change over time 21

Identifying Vulnerabilities: Use Non- Predictive Models (‘Exploratory Modeling’) • Model is used to explore (“what if . . .”) • Simultaneously takes into account context uncertainty, model uncertainty, and uncertainty about weights (multi- dimensional sensitivity analysis) • Objective: reason about system behavior ─ under what circumstances would a policy succeed or fail (‘scenario discovery’) • Basic idea: explore the consequences of the unresolved uncertainties, and which would make a difference • Requires a huge number of computer runs 22

Key element of Dynamic Adaptive Policies: A monitoring system with related contingency actions to keep the policy on track MONITORING SYSTEM: • Signposts and trigger values • Are we still on track? • Are corrective actions needed? • Do we need to implement actions earlier or later? • Is reassessment needed? 23

What Happens After an Adaptive Action is Triggered? Two possibilities 1. Restart the DAP process again, from the new situation (as shown on the DAP slide) 2. Identify promising pathways (i.e., sequences of policy actions) in advance (e.g., to avoid lock-ins) • We call this the Dynamic Adaptive Policy Pathway (DAPP) approach* • Based on triggers called ‘tipping points’, which are conditions at which a policy begins to perform unacceptably • Can be graphically represented as a ‘Metro map’ *M. Haasnoot, J.H. Kwakkel, and W.E. Walker (2013). “Dynamic Adaptive Policy Pathways: A New Method for Crafting Robust Decisions for a Deeply Uncertain World”, Global Environmental Change , 123: 485 – 498. 24 24

Example of an Adaptation Pathways Map • Several paths will satisfy the policy objectives • Some require many changes in policy; some require few • Some cost more than others 25

Example of an Adaptation Pathways Map • Several paths will satisfy the policy objectives • Some require many changes in policy; some require few • Some cost more than others 26 26