• C++ Programming for Financial Engineering
    Highly recommended by thousands of MFE students. Covers essential C++ topics with applications to financial engineering. Learn more Join!
    Python for Finance with Intro to Data Science
    Gain practical understanding of Python to read, understand, and write professional Python code for your first day on the job. Learn more Join!
    An Intuition-Based Options Primer for FE
    Ideal for entry level positions interviews and graduate studies, specializing in options trading arbitrage and options valuation models. Learn more Join!

Modelling Electricity Prices

Hi all,

I signed up specificaly to post here.
What I can say is that your approach is bad. If you are at uni professor will love your approach, if you are not, bosses will dislike it (I think).
What I learn about electricity market is that derivatives don't care of the spot. It's too risky, no one "knows" when spikes will occur. Remember a derivative desk is not looking to do prop trading, it just wants to lock the extra premium it charges to the client (which can be very huge). Like in the markovitz efficiency frontier, they are looking to lock the more money with the less variance. When you are you using spot you are adding too much variance.
If you want to price a hydro plant you can take a look to extrinsic valuation (replicating the plant with forward and calendar spread options), it will lead to a lower bound, however this lower bound is realisable for traders.
Of course the point of view is different if you are looking to sell or buy the plant. If you want to buy it, you will have to find a lower bound which you are sure your traders will reach. If you want to sell, then you maybe can take into account the spot price to get a very high upper bound but which can't be reachable by traders.

At the begining I was like you, thinking how it was complicated to model the spot and to integrate it into the trading strategy. However it will lead to an upper bound which will be very complicated to reach for traders.
 
Thank you for the precious time you've taken to answer my post.

As I think I mentioned in one of my previous posts, I have decided to model the residuals between a seasonality function and my time series. I have discussed about this approach with my teacher, and we both agreed that for the purpose of my project and due to the time constraints, the model was sufficiently good.
 
Nimarog said:
Hi all,

I signed up specificaly to post here.
What I can say is that your approach is bad. If you are at uni professor will love your approach, if you are not, bosses will dislike it (I think).
What I learn about electricity market is that derivatives don't care of the spot. It's too risky, no one "knows" when spikes will occur. Remember a derivative desk is not looking to do prop trading, it just wants to lock the extra premium it charges to the client (which can be very huge). Like in the markovitz efficiency frontier, they are looking to lock the more money with the less variance. When you are you using spot you are adding too much variance.
If you want to price a hydro plant you can take a look to extrinsic valuation (replicating the plant with forward and calendar spread options), it will lead to a lower bound, however this lower bound is realisable for traders.
Of course the point of view is different if you are looking to sell or buy the plant. If you want to buy it, you will have to find a lower bound which you are sure your traders will reach. If you want to sell, then you maybe can take into account the spot price to get a very high upper bound but which can't be reachable by traders.

At the begining I was like you, thinking how it was complicated to model the spot and to integrate it into the trading strategy. However it will lead to an upper bound which will be very complicated to reach for traders.
Click to expand...

I wouldn't say his approach is bad. It's just a very academic way of looking at the valuation.
 
To start I would go for a multiparametric mean reversion model that can be calibrated on mean levels of forward prices at screen. I have found it working quite well for derivatives pricing, VaR calculation and hedging purposes. More specifically you can try to implement mean reversion with jump diffusion.
Other statistical approaches can be found in the popular Weron textbook and a more complex semi-empirical model is by Roncoroni (2006) "Understanding the fine structure of electricity prices". A cohemprensive review is in the book by Fiorenzani
To learn fast the techniques I can suggest a dedicated workshop on energy quantitative analysis and modelling, the EQuanT Bootcamp, among the tutors will be Weron and Fiorenzani themselves.
For further insight on the topic can also join a dedicated community on Linkedin and get replies from energy quant professionals.

Hope it helps!
 
A few years later with more experience in the Energy industry - I would actually say now that which forecasting method you use depeneds on WHAT electricity price you are trying to forecast.

There are many different markets in Power, e.g., Day Ahead (DA) market, Hour Ahead (HASP), Fifteen Minute (FMM), and Real-time (RT) or 5-minute market. In general, vol increases the closer you get to t=0 (not strictly the case). So for forecasting DA prices you can use something like ARIMA model with added seasonality.

RT models will require something with higher fire-power and will most likely have higher forecasting error as spikes are unpredictable.
 
Back
Top