How to build a robust model with almost no data
After doing the OpenOil training I had to choose a project I was interested in modeling. The joint venture between YPF and Chevron to exploit shale gas and oil in Vaca Muerta (South West of Argentina) was the obvious option. It had been all over the news, since the U.S. Energy Information Agency (EIA) had announced in 2013 that Argentina had one of the world’s largest reserves of shale gas and oil.
When I started to gather data, I realized that there was very little information, compared to the OpenOil online trainings I had watched. In the videos you could see how easy it was to find almost all relevant facts by looking into financial statements or corporate presentations. My case was nothing of the sort. Both YPF and Chevron published very little information relevant for the model, usually not more than a brief paragraph in the 20F submitted to the U.S. Securities Exchange Commission (SEC). The Energy Ministry of Argentina and Neuquen (the province where Vaca Muerta is located) didn’t help much either. I could only find aggregate production by month and year, but the rest was either too confusing for me to make sense of it, or it related to the whole of the Vaca Muerta region (instead of specifically to the project I was modeling). In essence, after a lot of time spent looking for data, I realized that other than the actual annual gas and oil production for years 2014 and 2015, my only really useful piece of information was just one slide of a corporate presentation of the project.
The slide merely gave totals for the project that would last 35 years: total investment of USD 16,000 million; total operating cost of USD 9,000 million; 1,500 new wells; 750 million barrels of oil equivalent (BOE); USD 8,500 million royalties for the province, and the peak rates for oil (50,000 barrels per day) and for gas (3 million cubic meters per day). After that, with the help of Iliusi Vega from OpenOil and Aleph (see her blog), I was able to find local prices for gas and oil in some financial statements, although I then had to understand the details and calculations for the subsidies for both gas and oil.
With the help of OpenOil’s expert Alistair Watson, we constructed a first sketch of the model, trying to simulate the total gas and oil production to reach a total of 750 million BOEs based on the peak rates and the actual production for years 2014 and 2015. With this data and the value of royalties and the price of gas, we were able to estimate the oil price envisioned by the company when the joint venture took place. As for wells and development costs, there was little (and contradictory) information about the number of drilled wells, and how many were vertical and horizontal. Therefore, we only used bulk numbers of development costs based on some company press releases about the investment during the 2013-2014 pilot project and the investment expectations within the next years.
It was then time to construct the model, for which I used a template from another oil project. Being a lawyer it was indeed challenging to understand the model, its different sheets, relationships, formulas, and to adapt the template to the specifics of my project: adding gas production, special gas subsidies, different oil subsidies, and preparing different scenarios (depending on the time the peak was reached, the availability of subsidies, gas and oil prices, etc).
When I got to Berlin’s Sprint, my stage 1 model (pre fiscal cash flows) was almost done, and I was half way through stage 2 (the fiscal regime). After really intense – but fun – days of work, feedback from experts and peer reviews with other learning-modelers, my model was finally done.
Seeing the dashboard, which summarized all of the model’s results and allowed me to instantly see the impact on production, cash flows or internal rate of return upon any change to parameters… it was simply magical. Given the many factors influencing my model (peak rate and duration, availability of subsidies, prices, operating and development costs, all applicable both to gas and oil), OpenOil’s expert Alistair Watson taught me how to create scenario and sensitivity analyses, where we could model changes to many of the parameters all at once. This way we could see how results changed from the base scenario, to the price and production reality of 2016 and then to predict how results would change under more favourable scenarios, or upon changes to the fiscal regime (including the repeal of subsidies).
Upon returning from Berlin, I got in touch with people from YPF who informally agreed to look at my model and let me know if any of my parameters was way out of line. I was happy to find out that they broadly validated the model, in spite of the development cost assumptions not related to the drilling of wells (I didn’t have enough information to model this). They suggested to fix some parameters related to the fiscal regime and the peak duration, both of which took a few minutes to change. They suggested different parameters for cost reduction (in an optimistic future scenario). It took me just a few seconds to change those values.
All in all, this was a great learning experience, that proved the usefulness of modeling, which goes beyond the extractives industry. It allowed me to ask the right questions, to question basic assumptions or claims which contradict the model, and I hope that it will provide a useful tool now that the fiscal regime and the cost of subsidies of the gas and oil sectors seems to be under discussion.
The financial model and narrative report are available here.