The IEA's method of prediction seems to bring supply up to demand, by assuming price changes and additional investments to provide the needed supply. The increased prices slightly affect demand, so the process is iterative.

These are the words of the IEA on methodology:

Matching oil supply and demand

Total oil demand is the sum of regional oil demand, world bunkers and stock changes. Equilibrium between oil supply and demand in the World Energy Model results from an iterative process. First, projections of oil supply from countries closed to international investments are run separately under a policy-driven scenario of national investments. Second, projections for the group of countries open to international investments are run under a given price scenario.

Third, the resulting total world oil supply is then compared to the results of the oil demand module under the same price scenario. The process of adjusting the price scenario and the investment policies of closed countries is repeated until an empirical convergence is found between closed countries’ capabilities and investments policies, open countries’ market-driven supply potential and the price elasticity of oil demand in the WEM regions.

The world has just completed a giant live experiment by trying to grow the economy against limited crude oil supplies at around 74 million barrels/day. Result: bank failures, company bankruptcies, airline closures, a car industry at its knees and many rescue packages which mean new debt. Assuming we come out of the current crisis, who would want to repeat that exercise? When will we repay debt at all?

Thanks for bringing up this excerpt. I had also to do with modelling in my earlier life. However these models dealt only with natural parameters, so it had a fairly narrow and "predictable" range of results.
But the IEA model has to deal with quite a few "unpredictable" parameters, e.g. politics, so they have a big array of srews they turn manually according to their personal assumptions. So simply depending on these manual adjustments the model can produce an huge array of results.
It is a pity that they don't provide more transparency by publishing all details (source code, data etc.) on the modelling process.