Debika -

Any kind of energy forecast 25 years into the future has almost no chance of being reasonably accurate, due not only to the usual variance and bias issues, for both the input data quality and the model, but also because of breaks that will occur in the time series, which are largely unknowable ahead of time.  One may test, if useable data are available, to see how  accurately a forecast would have performed if used previously, but that does not tell you what new impacts may occur in the future due to technical developments, politics, or anything else. 

Remember that the difference between prediction for a cross sectional survey, and a time series forecast, is the general unavailability of current data in a forecast:

 https://www.researchgate.net/publication/275365705_When_Prediction_is_Not_Time_Series_Forecasting

Short-term forecasts would generally be expected to be more successful, as the model can be revised when needed, and data discontinuities/time series breaks have fewer chances of occurring.  This would be especially true should they both forecast monthly.  For a short-term forecast you might want to use exponential smoothing, which puts more emphasis on more recent data, but is still capable of incorporating seasonality. For a long-term forecast, you may do well with ARIMA, but you should be careful not to 'overfit' your model in either case. 

(See  https://www.researchgate.net/publication/220151263_Ockhams_razor_empirical_complexity_and_truth-finding_efficiency

 - pages 1 and 2, especially the comment on 'overfitting' on page 2.)

Also, when modeling price, keep in mind that that is a ratio of two variables.  For the cross sectional energy establishment surveys on which I worked for many years, I looked at each variable in price separately.  Otherwise consider how one price may represent a much smaller volume than another, and should not be equally weighted. I am not sure of your purposes for your application, but I think you should keep this in mind. 

The US Census Bureau, and others, have done long-term forecasting which you may want to research.  But for energy data, you may want to first study what has been done on the National Energy Modeling System (NEMS), starting with the following link:

 http://www.eia.gov/oiaf/aeo/overview/

This is on the US Energy Information Administration's (EIA's) website, which may contain other useful information for you, specific to diesel fuel, or something related.  However, I know that much of the NEMS input data are less accurate than assumed when using that system, and I'm not sure it isn't overparameterized, and there are always those pesky time series breaks to consider. 

You will generally see forecasts which are published to a ridiculously large number of digits, when the model variance alone would make most of the digits published unjustifiable. 

The EIA website has a huge amount of information available, but most of it is far less accurate than its presentation would lead you to believe. Still, there are technical notes, tables of relative standard errors, and other information available for you to use to assess the usefulness of much of what you may find there.

So caveats and metadata/paradata are in order.  An uncertainty analysis is very important, perhaps especially for long-term energy forecasts. 

As part of your evaluation, I suggest substantial use of graphics.

Best wishes - Jim

PS -  One somewhat bright note is that for a price, which is revenue divided by sales volume, if you overestimate one, you tend to overestimate the other.  Similarly, if you underestimate revenue, you tend to underestimate sales volume. The net effect can be that price tends to be estimated somewhat more accurately than sales or revenue, or at least it isn't worse than both.  

Article Ockham’s razor, empirical complexity, and truth-finding efficiency

Research When Prediction is Not Time Series Forecasting

http://www.globalpetrolprices.com/articles/51/