Skip to content

Another Victory for the Behavioral Economists

A recent working paper shows that insights from behavioral economics help explain consumers’ choices in a demand response program.

Richard Thaler won the Nobel Prize in Economics last month, making him the third behavioral economist to earn this honor in the last 16 years (Daniel Kahneman, of Thinking Fast and Slow fame, won in 2002 and Robert Shiller won in 2013). The members of the Nobel committee must think the behavioralists are on to something.

Richard Thaler, 2017 Nobel laureate

In a nutshell, behavioral economists bring insights from psychology to economics. A recent paper by James Gillan, a researcher at the Energy Institute who is finishing his PhD this spring, highlights that electricity markets are not immune to the influences of human psychology.

Gillan’s study takes on a sacred cow in economics: the law of demand. The law of demand holds that when the price of something goes up, people will consume less of it. You might have a verbal memory of your first economics professor reminding you that, “demand curves slope downwards.” This law is basically to economics as the law of gravity is to physics.

A demand curve

Belief in the law of demand for electricity has led a number of us to advocate for dynamic pricing (here and here). The cost of producing electricity can vary wildly over short time periods, but the vast majority of retail customers face rates that do not vary at all over time. Why not charge customers high retail prices when the electricity system is pushed into overdrive, and, conversely, lower prices at other times? If customers reduce demand in the face of high prices, the system operator may be able to balance the system with fewer expensive and dirty peaker plants. Over time, that means we’re building fewer power plants.

Gillan conducted a novel and carefully designed field experiment to test this logic. He partnered with an unnamed demand response provider (who I’ll call DRP) that recruits residential customers who are willing to receive text messages that incentivize them to reduce their consumption during 1-hour event windows.

Gillan worked with the DRP to randomly allocate a subset of their new customers to a 3-month holding period during which they didn’t receive any event notifications. These customers were the control group. The remaining customers were exposed to between 1 and 3 events per week, called between 11AM and 10PM.

Here’s the really amazing part, though. The DRP varied the incentive during these events from $0.05 per kWh to $3.00 per kWh. You can think of these as adders to the amount the customers owed the utility, which is on average $0.16 per kWh. In particular, consumers had a baseline and were rewarded if they went below it during an event and penalized by the same amount if they went above.

As someone who has spent a lot of time trying to convince various partners to run experiments, I am really impressed that Gillan and the DRP pulled this off. Kudos also to the California Energy Commission for supporting the study. The experiment yielded some super interesting findings.

Consumers Do Reduce Consumption When the Price Goes Up

The first result in the paper validates our basic Econ 1 models. During the event windows, the customers in the treated group consumed on average 12 percent less than the customers in the control group. So, higher prices imply less consumption. Check.

It’s also interesting to see these reductions because the DRP’s customers were given very little time to react. Under many utilities’ Critical Peak Pricing programs, for example, customers are alerted the day before a price change, but the DRP sent out the text messages one hour before the event. In California, real-time prices are a lot more volatile than day-ahead, so we’d ideally like customers to be exposed to real-time prices as they reflect the most up-to-date information about the electricity system.

Consumers Are Extraordinarily Insensitive to the Price Level

Unfortunately, things stop looking quite as good for Econ 1 if you zero in on the responses to different events. Recall that the price bump varied by a factor of 60: from 5 cents to 3 DOLLARS. Gillan shows that the reductions did not vary by much at all – around 11 percent reductions for 5 cents and 13 percent for 3 dollars. That means people were behaving as though they were really insensitive to the price, which is hard to square with the fact that they reduced by 12 percent across all the events.

Gillan attributes this insensitivity to the price level to what the behavioral economists call “scope neglect.” Previous research has shown, for example, that people are willing to donate the same amount of money to save 2,000 endangered birds as 200,000. Yes, I care about endangered birds, but I have trouble thinking about the difference between 2,000 and 200,000. In the demand response context, Gillan’s results suggest that people pay attention to the fact that there’s a change in price, but not the size of the change.

Gillan also finds that people responded less to a separate set of messages that had no price change – so they’re not simply reducing because they see a message. It seems that the hurdle lies in translating the price displayed to the possible savings.

Automation Technologies May Substitute for Neglectful Consumers

A random subset of the treated customers was also encouraged to install smart thermostats or smart plugs in their homes. These devices let the customers ignore the text messages about the events as the smart devices would turn off the connected equipment for them.

Gillan finds that installing the devices led to considerably larger reductions, suggesting that consumers are willing to reduce even more when demand response is automated. The particular devices in Gillan’s experiment didn’t let consumers program different responses to different price levels (and his results on consumer’s insensitivity to the size of the price change are equally convincing for consumers with or without smart devices). But, you could imagine that more sophisticated devices would allow this, which could lead to a more traditionally shaped demand curve.

In the end, I think it’s possible to claim victories for both economics and psychology – two half-full glasses. While it’s true that Gillan’s results don’t suggest a nicely shaped downward sloping demand like we draw in Econ 1, the basic logic behind dynamic pricing holds – consumers reduce consumption even when they learn about the price hike one hour before the event. And, his results highlight the importance of paying attention to insights from psychology, just as the Nobel committee has been doing recently.

Categories

Uncategorized

Tags

Catherine Wolfram View All

Catherine Wolfram is the Cora Jane Flood Professor of Business Administration at the Haas School of Business, Co-Director of the Energy Institute at Haas, and a Faculty Director of The E2e Project. Her research analyzes the impact of environmental regulation on energy markets and the effects of electricity industry privatization and restructuring around the world. She is currently implementing several randomized control trials to evaluate energy efficiency programs.

9 thoughts on “Another Victory for the Behavioral Economists Leave a comment

  1. “Victory for Behavioral” makes it sound like traditional economics is losing. Behavioral economics expands the tools and the domain of economics, right?

    As Julian Huxley and Jan Tinbergen’s brother Nikolaas argued, we should recognize levels-of-analysis in scientific explanation. The downward sloping demand curve is derived at the level with no transaction costs, in particular decision costs. Behavioral economics is at another level (see especially Dixit’s “The Making of Economic Policy” on the 1st, 2nd, and 3rd-best levels). It is up to the analyst to decide which level is appropriate for a particular problem.

    Kahneman and Tversky understood that rules-of-thumb may be rational in the face of decision costs and sought to find situations wherein those rules could be improved, not to advocate that all behavior be analyzed at the level of Thaler’s “Misbehaving.”

    In particular (and as suggested by Alan), it may not be rational for households to revise decision-making shortcuts where the stakes are small. Start with commercial and industrial customers, and demand aggregators for residential consumption.

  2. Repy to David Jacobowitz:

    Perhaps we should take this to private correspondence, but here is a quick response.

    Economists are well acquainted with state-space modeling and feedback issues. I myself spent the first ten years after graduation focusing on the development of robust filtering, smoothing, and control methods for nonlinear economic systems for which the dynamic and measurement relationships are only partially understood, hence for which model misspecification is a primary concern.

    However, the instability issues studied in our braided cobweb paper are not the usual short run problems handled by power system operators either at the transmission level or the distribution level but rather an increasing volatility in power prices and power levels that can build up over successive days of **end-to-end** power system operations with dynamic price retail contracting that involves one way communication of wholesale-based prices to retail-owned and operated devices. You are correct that these issues can be handled, but as discussed in our paper it would seem to require a demand response program that is based on two way communication between a distribution system manager and smart price-responsive devices, such as the types of bid/offer based designs being proposed by “transactive energy system” (TES) researchers. This is the type of design under study in my current DOE/PNNL projects.

    At the wholesale level, in regions with ISO/RTO -managed wholesale power markete (MISO, PJM, ISO NE, ERCOT, SPP, CAISO, NYISO), the commitment, dispatch, and price solutions for the day ahead market are determined by means of bid/offer-based SCUC/SCED optimizations subject to physical feasibility and reliability constraints. So a key issue is whether a distribution system should likewise be centrally managed by some form of distribution system operator (DSO) or whether a more decentralized TES design would work better.

    Ok, way more than I intended to get into here!

  3. I have been somewhat skeptical of dynamic pricing without enabling technology, and this study actually points to one of the reasons. When load got high in the past, before dynamic pricing, there would be public appeals of various sorts (Flex Your Power being the latest incarnation), and demand would drop somewhat due to the appeals causing some folks to make short-term changes (not doing laundry, turning A/C thermostat up, etc.). It has been my hypothesis that dynamic pricing includes the demand drop from the “public appeal” effect so that demand reduction is therefore overstated (relative to what would happen in the real world without dynamic pricing) when measured against a baseline of demand on a more normal day without a dynamic pricing event or public appeal. This behavioral economics experiment shows that what is important to customers is the signal that they need to do something – not how much they are rewarded or penalized for changing behavior – in other words, a public appeal with a little bit of money to back it up.

  4. I think this just indicates that the study targets the wrong sector of consumers. For many householders this is really too small a savings or cost to spend much time on. The changes in cost are a very small part of a total household budget.
    Market based demand response will probably be much more effective in the commercial sector—office buildings, hotels, store chains,, and the like. In those businesses a modest amount of costa d effort could result in real savings—enough to justify the cost of managing energy.

  5. Interesting cautionary commentary on dynamic pricing. The study below uses analysis and computational experiments to highlight another critical issue that can arise for dynamic-price retail contracting in centrally-managed wholesale power markets based on a two-settlement system, even **with** smart automated household metering: namely, one-way communication of prices to devices does not properly take into account the possibility of induced dynamic instability in end-to-end power system operations over time.

    Auswin G. Thomas and Leigh Tesfatsion, “Braided Cobwebs: Cautionary Tales for Dynamic Retail Pricing in End-to-End Power Systems,” Economics Working Paper No. 17028, Department of Economics, Iowa State University, Ames, IA, 2017. http://www2.econ.iastate.edu/tesfatsi/BraidedCobwebsWP.ThomasTesfatsion.pdf

    • Dr. Tesfatsion,

      Thanks for posting your interesting paper. I know slightly more about controls than I do economics, but not very much about either. The cobweb analysis sounds like economists have discovered feedback in stateful systems. There is an enormous amount of engineering expertise in control theory for managing the behavior of complex systems with linear and nonlinear feedback; would it not be possible for the system operator to take the transfer function of the price-responsive load into account when sending back the price- or control- signal, thus avoiding or quickly dampening any price/quantity oscillations? Or, one could imagine a system by which the system operator could have rights to modify the transfer function of the customer’s automatic systems, by conveying additional parameters (like PID coefficients) along with the price.

      Also, curious if/why this is more problematic for price responsive demand than for price responsive supply which we’ve had for some time? If you have generators that take time to respond to a new price signal, one might expect to see oscillations. Has that been observed? I believe, for example, that ISOs run various simulations from minutes to hours in advance in order to detect when they should pre-commit long startup resources that may/will be needed much later. I’ve imagined this as an effort to avoid price spikes and the need for sudden significant redispatches, but it could also be in the service of stability more generally.

      That is, is this a real fundamental problem, or just news that grid operations software will have to be that much more sophisticated?

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: