U.S. Water Prices: a Machine Learning Approach

Many communities face issues involving water scarcity, and the problem is expected to continue to grow. Expanding the allocative toolkit for this resource will help relevant stakeholders (municipalities, farmers, environmental managers, etc.) better deal with these challenges. One possible solution to mitigate the effects of water scarcity is by harnessing market mechanisms to help water flow to its most valuable use. Market tools will not be a panacea for all the water woes inherent in increasing scarcity, but they can help increase allocative efficiency. Water markets exist in a limited capacity in the United States, particularly in the West, yet these markets do not have a high level of participation or transparency. Looking elsewhere in the world we see that water markets have been successfully developed, and implemented, with the most robust example being in Australia. To better inform market development in the U.S., a random forest pricing model was constructed based on the Australian experience that used a combination of market data and bio-physical inputs to predict water prices as the output. The selected inputs were chosen so that the same data was available in the potential U.S. locations.

The random forest model is trained over the period 2007 to 2015 to predict the cash value of water in the Australian market. Model performance was compared to actual cash prices for water for validation and predictors’ variable importance was explored. The inputs include commodities that are water dependent (such as cotton) as well as an environmental variable (precipitation). Following a successful method verification for the Australian market, the method was transferred to the U.S. (the Guadalupe River Basin in Texas) where the model was run using identical inputs with U.S. values. After volumetric and currency conversions were applied, the resulting output is a theoretical price of water in the U.S. The Guadalupe River Basin has some options styled water trading, so offers the opportunity—though limited—for validation of the transferred model. These valuations can help gauge the level of interest of potential market participants at different price points and may inform initial pricing in a new water market. Additionally, combining data types from different sources—in this case market and bio-physical—begins to shed light on novel approaches to developing data driven solutions by creative data fusion.