Introducing the Hivemind El Niño Prediction Market
The market aggregates the judgment of experts to generate a probabilistic forecast of the global weather phenomenon.
El Niño-Southern Oscillation
Under normal conditions there is a temperature gradient in sea surface temperatures across the tropical Pacific with warmer waters in the west, near Indonesia, and cooler waters in the east off the coast of South America. This gradient is maintained by winds blowing across the ocean from east to west. These winds are themselves driven by the temperature gradient, with air rising over the warm pool, travelling eastwards aloft before descending in the east.
During an El Niño event these winds weaken causing the temperature gradient to diminish, which further weakens the winds. The eastern tropical Pacific becomes abnormally warm. This state can persist for several months before it is reversed by the oscillatory dynamics of the coupled ocean-atmosphere system.
In contrast to warm El Niño events, La Niña events are caused by a strengthening of the winds and of the east-west gradient in sea surface temperatures. During such an episode the central and eastern Pacific become unusually cool.
El Niño events typically occur once every two to seven years and the irregular cycle is called El Niño-Southern Oscillation (ENSO). It is the single largest cause of year-to-year variability in global temperatures.
Impacts of El Niño and La Niña events
The Pacific is the world’s largest ocean and surface temperatures rising by a couple of degrees Celsius across a large portion of the tropical ocean, as happens during an El Niño, has repercussions for weather systems across the globe.
During an El Niño there can be increased rainfall on the west coast of South America and North America. This is mirrored by a heightened risk of drought on the other side of the Pacific, in Indonesia and Australia. Other impacts include a reduction in the frequency of Atlantic hurricanes and an increase in the risk of failure of the Indian monsoon.
Similarly La Niña events affect weather across the globe, increasing the risk of drought in parts of the United States, for example.
The effects of ENSO on global weather makes the phenomenon of interest to many constituencies including farmers, commodity traders and insurance companies.
Expertise counts, but how do you count expertise?
Prediction markets: A tool for aggregating judgment
Prediction markets share features with both traditional gambling and financial markets but their primary purpose is to obtain information. Information discovery is a side effect in other markets whose primary purpose might be the allocation of capital or the transfer of risk but it is the main objective of a prediction market.
The assets in a prediction market are typically contracts that pay out 1.00 unit if a specified event occurs. These assets are traded, either directly between participants using a double auction, or through a market maker that is always willing to quote buy and sell prices. Assuming the assets are trading at fair value, the market price of a contract is the implied probability of it paying out, that is the probability that the outcome covered by the contract will occur.
Last year Winton sponsored a prediction market competition in which two dozen teams from British universities took part. They had to predict what the monthly average temperature and rainfall for the U.K. would be. This market used a market making algorithm to ensure liquidity and was described in a previous blogpost.
This year Hivemind is going to follow-up with a prediction market for ENSO. The target of the market will be the monthly value of a variable known as NINO3.4 SSTA.
NINO3.4 SSTA: The S&P 500 of tropical climate
Every El Niño event is different but their defining characteristic is anomalous warming in the central and eastern tropical Pacific. The state of ENSO is often summarised using a single number: the average sea surface temperature in a region of the Pacific known as NINO3.4. This number is usually expressed as an “anomaly”: the difference between its value and the average value at that time of year in a baseline period (e.g. 1981–2010). The sea surface temperature anomaly in the NINO3.4 region is abbreviated as NINO3.4 SSTA.
The NINO3.4 SSTA is analogous to the S&P 500 stock market index: The S&P500 index doesn’t tell us what is happening to individual stocks but by averaging over 500 of the largest US companies it provides a useful summary of the state of the US stock market. Similarly, NINO3.4 SSTA cannot fully capture the temperature structure of the Pacific ocean but it provides a concise indication of the phase of ENSO. In particular, values of more than 1°C imply El Niño conditions, with the anomaly exceeding 2°C during large events. Negative values indicate La Niña episodes.
Design of the market
The Hivemind ENSO market runs on Hivemind’s Agora platform for hosting prediction markets. In the ENSO market there is actually a separate market for each month. The outcome space for each market partitions NINO3.4 SSTA into intervals of 0.1°C covering the range -4°C to +4°C with the left and right intervals being open (covering all outcomes below -4°C and above +4°C respectively). Participants are endowed with credits for each monthly market and credits are not transferable between months.
Participants can define contracts covering one or more intervals. Once a contract is defined Agora will quote a price at which the participant can buy the contract. Each contract pays out 1.00 credit at settlement if it contains the interval covering the actual value of NINO3.4 SSTA.
The market for a given month remains open for trading until the last day of that month. Once the actual value of NINO3.4 SSTA is published the market is settled with contracts covering the correct outcome being converted to credits. The credits then become lottery tickets in a lottery for real cash awards. The probability of a participant winning a cash award is equal to the fraction of the final credits they hold in that month’s market. Hivemind is sponsoring a monthly cash award of £1000 but is seeking other organisations who would like to sponsor further cash awards.
Participation
Participants do not have to pay to take part in the ENSO market. In order to keep the expected cash awards high enough to incentivise informed participation the total number of participants is limited. To join the market participants must show evidence of relevant expertise for ENSO prediction. This could be a background in seasonal forecasting or more general experience in time series prediction using statistical or machine learning methods.
Unlike the previous climate market, in which the cash prizes were awarded to the institutions of the best performing teams, the ENSO market will be open to individuals who will be eligible for the cash awards.
Further Reading
To learn more about ENSO and ENSO prediction see Columbia University’s International Research Institute ENSO Resources
A historical record ENSO indices including NINO3.4 SSTA is available from NOAA.
If you work for an organisation that would like to co-sponsor the market or if you have relevant expertise and would like to participate please contact Mark Roulston.
