Managing uncertainty in decision making: What can we learn from economics?

Community member post by Siobhan Bourke and Emily Lancsar

Siobhan Bourke (biography)

How can researchers interested in complex societal and environmental problems best understand and deal with uncertainty, which is an inherent part of the world in which we live? Accidents happen, governments change, technological innovation occurs making some products and services obsolete, markets boom and inevitably go bust. How can uncertainty be managed when all possible outcomes of an action or decision cannot be known? In particular, are there lessons from the discipline of economics which have broader applicability?

Emily Lancsar (biography)

While uncertainty is often discussed alongside risk, a fundamental difference between uncertainty and risk is that risk involves events with known probabilities (or probabilities based on reliable empirical evidence), whereas under uncertainty probabilities are unknown and reflect an individual’s subjective belief concerning the likelihood of a given outcome. Given the subjectivity, that likelihood can differ from person to person. It can also involve a perceived zero probability in the case of unforeseen events (or ‘unknown unknowns’).

We highlight three approaches from economics that have broad value in managing uncertainty, especially for helping decision makers in taking uncertainty into account: expected utility theory, hedging, and modelling. A common strength of these approaches is that they explicitly consider uncertainty rather than ignoring it.

Expected utility theory

Expected utility theory provides a useful approach to choice under uncertainty. It helps decision makers think about different options in terms of the probabilities of those options occurring and to rank them.

In particular, expected utility is the utility an individual (or some aggregate of people) is expected to obtain under different circumstances or ‘states of the world’. It is calculated by taking the weighted average of all possible outcomes with the weights reflecting the probability that a specific event will occur. Decisions are made to maximise expected utility.

Expected utility theory is underpinned by an assumption of rationality. Extensions and alternatives to expected utility theory have been developed to account for lack of rationality in human behaviour and cognitive biases. A notable example is prospect theory (developed by Kahneman, a Nobel prize winner in economics, and Tversky, 1979) which describes how people choose between uncertain alternatives. It highlights the importance of reference points and recognises that individuals may view potential losses differently from gains.

Hedging

Hedging can be used to manage both risk and uncertainty. Hedging is like insuring against an uncertain adverse outcome by offsetting potential losses associated with uncertain events by gains in other investments. This doesn’t prevent the negative (uncertain) event occurring, but reduces the adverse impact of that event, should it occur.

This can be achieved by taking a portfolio approach such that decision makers diversify across investments or courses of action. As economists are fond of saying ‘there is no free lunch’; reducing the adverse outcomes of an uncertain outcome usually comes at a cost of a reduced rate of return in certain times for increased survival in adverse times.

Modelling

Modelling―especially systemic representations of complex real-world scenarios and simulations that account for the subjective probability of such scenarios occurring―provides a way of assessing existing (and projecting future) inputs and outputs and systematically testing the impact of policies in ways that include and account for uncertainty. Modelling is used not only in economics, but also in a wide range of other disciplines and fields as attested by multiple contributions to this blog (see https://i2insights.org/tag/modelling/).

What economics particularly brings to modelling is a basis in economic theory which is fundamental both to the building of the models and interpretation of the results along with focus on causal relationships and the explicit inclusion of uncertainty. A specific example is the development of decision analytic models to explore the cost effectiveness of health technologies (including drugs, devices, services, etc.) to inform government resource allocation decisions. Such models are used to reduce uncertainty regarding the question of value for money and guide government investment decisions.

Do you have other lessons from economics to share? Have you applied economic approaches to uncertainty outside the discipline? Are there issues in dealing with uncertainty that economists could usefully apply themselves to?

Reference:
Kahneman, D. and Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47: 263-291.

Biography: Siobhan Bourke PhD is a research fellow in the Department of Health Services Research and Policy, Research School of Population Health, The Australian National University in Canberra, Australia. Her research has focused on economic evaluations, orphan drugs policy, health policy and patient reported outcomes.

Biography: Emily Lancsar PhD is a Professor and Head of the Department of Health Services Research and Policy in the Research School of Population Health at The Australian National University in Canberra, Australia. Her broad research interests are in health economics, with a particular interest in understanding and modelling choice, preferences and behaviour of key decision makers in the health sector. Emily also leads the recently created ANU Health Policy Lab, premised on engagement and co-creation of research with policy makers and practitioners with the ultimate aim of contributing to improved health and wellbeing.

Siobhan Bourke and Emily Lancsar are members of blog partner PopulationHealthXchange, which is in the Research School of Population Health at The Australian National University.

14 thoughts on “Managing uncertainty in decision making: What can we learn from economics?

  1. Always nice to see discussion of techniques for managing uncertainty. I’m curious in this case as to why these three were selected, and what others were perhaps considered before describing these ones?

    In a sense they are used for completely different purposes. Expected utilities provide a way of collapsing a probability distribution down to a point, to make it easier to choose between alternatives. Hedging takes an original decision problem and broadens the scope of alternatives to achieve “better” outcomes. Modeling helps reduce uncertainty by combining and structuring sources of information (including uncertainty about them).
    The relationship between the three is not clear, though they do seem to hint at a broader framework with gaps that have not yet been explored?
    All three could be used together, e.g. selection between alternatives based on expected utilities within a real options framework which involves investing in a portfolio of modelled short and long term actions, including insurance.

    One criticism – I prefer avoiding the idea of “lack of rationality in human behaviour”. If anything, the work on heuristics has shown that they can be fit for purpose, and are underpinned by their own rationality, i.e. reasoning. Even though the term can be used in a technical sense, it is too easily misinterpreted as privileging a particular method of decision making over many others.

    • Thanks Joseph for your thoughts. We chose these techniques to demonstrate three different approaches. Agree can be used in isolation or in combination. We used rationality or lack of in a technical economics sense but fair point, especially given interdisciplinary audience, thanks for raising it.

  2. Thanks for this nice blog, and great to connect with others too. Just one more dollop of 5 cents below, and slightly mischieveous 🙂

    Economics has certainly been good at helping us make decisions: recognising trade-offs and making choices, if we agree on the payoffs, conditioned to risk and uncertainty. This is certainly where other disciplines can draw upon economics, and many have. However, the economics field in general can learn from other disciplines too.

    Economics has been pretty hopeless in its modelled projections of future states of the world, from macroeconomics to climate models. This may be heresy coming from an economist who often builds models! Most orthodox economists build models that are ‘static’ in the sense that statistical equations are fitted tightly to historical data. That’s fine if we are intent on predicting the past, and much of science is. The problem is when we use such models to make predictions about future. That is a like driving while looking in the rear-view mirror!

    An alternative approach is dynamic modelling (e.g. system dynamics, agent-based modelling) where the phenomena under study are represented in differential equations, and generating real-world complexity, such as interactions, feedbacks, tipping points, phase transitions. These kinds of models can be made to be ‘glass boxes’, interactive, and are superb learning tools – not intended to be crystal balls. They keep us modest. Within a dynamic approach scenario analysis can also become more realistic, and we can then appropriately use hedging and expected utility theory etc. So, modelling and decision making, like driving, should be a dynamic process where we update and respond. Beware however, the mainstream won’t be happy!

    As economists increasingly move into health, environment, social policy etc, then an interdisciplinary approach may be best to help make better decisions. The underlying theory could best be driven by subject matter experts, modelling codified into mathematics recognising change, and economics can layer-onto these models its approach to valuation and making decisions. But, at least let’s try and move beyond static models. Best to keep our eyes wide open when driving, and our hands on the steering wheel.

    • Thanks Kenny for your insightful comments, which overlap with Roman’s comment. We agree re the need for further use of dynamic models especially in complex interdisciplinary research. Interested in your thoughts on the types of policy/research questions for which more static models may remain fit for purpose and when dynamic models are essential?

      • My thoughts on this:
        – The primary weakness of static/empirical/regression model is that it is tied to the data used to develop it. It’s therefore still useful for prediction whenever conditions are “sufficiently similar” to the original data. Conditions may still be “sufficiently similar” even in cases where they have changed substantially, e.g. because the model then provides a useful reference case, or simply a preliminary estimate for further work to improve upon. Static models also have a long history of being used to tackle “why” questions.
        – Conversely, dynamic models are essential when conditions are known to have changed, where explanation of mechanisms is important, or where understanding of how a system changes is more important than simply predicting an end state.

        • Thanks for your thoughts, Joseph. Understanding the ‘how’ as well as the ‘why’ increasingly important.

  3. Thanks Bonnie for noting scenario analysis as an approach to accounting for uncertainty which we agree, can be powerful, particularly when we have some a priori expectations about different scenarios that could arise. We would be interested in your experience in choosing which scenarios to model.

    • I’ve had some experience exploring climate change scenarios (linked with historical environmental data) in workshops. The process is highly engaging and you get a good buy in from those involved in the process. Discussions can be very animated because a narrative is a powerful communication tool and a way of exploring assumptions that different people make.

      In my sustainability teaching, my Master students were constrained to describing desirable (a normative activity) scenarios to inform policy. Many students found this difficult – maybe a naturally dystopian disposition. But it did help to 1) work backwards to see what needed to be put into place to achieve that future and 2) applying a range of scenarios allowed them to determine if their proposed policy options were resilient to a range of futures.

      Plus, it’s fun 🙂

    • Like modeling itself, scenario analysis is a rather diverse field.
      I was involved in one paper that tried to give an overview of different techniques, specifically focusing on the connection to uncertain futures and decision making under uncertainty:
      Maier HR, Guillaume JHA, van Delden H, Riddell GA, Haasnoot M, Kwakkel JH (2016) An Uncertain Future, Deep Uncertainty, Scenarios, Robustness and Adaptation: How Do They Fit Together?. Environmental Modelling & Software 81 (July): 154–64. doi:10.1016/j.envsoft.2016.03.014
      I personally find inverse/bottom-up methods quite powerful – starting from outcomes and mapping back to what assumptions about the system and drivers can result in those outcomes.

  4. Trouble with economics is that so much of it is based on regression analysis. Good for a historical perspective, but failing to address underlying dynamics. In a world with shorter lead times, little or no buffer capacity, and quicker feedback loops, the traditional economics approach is no longer good enough. Especially when you add on top of that most peoples ignorance and understanding of lags, delays and dead times. That’s why the mess we are in and getting deeper. We need to understand and factor in the underlying dynamics and momentum of change, that’s why System Dynamics, not just Systems Thinking, is the only approach that addresses these. Many profess Systems Thinking expertise, few have System Dynamics expertise. System Dynamics is where Systems Thinking rubber hits the road. Our modelling needs to get real.

    • Thanks for your thoughts, Roman. Regression analysis is one but not the only analytical approach used in economics and we agree that accounting for underlying dynamics and momentum for change is important. System dynamic models provide a great opportunity to do that, particularly for complex systems, and is one approach we are currently pursuing.

Leave a Reply to Emily Lancsar & Siobhan Bourke Cancel reply