Power trading - forecasting
Hey all. I'm trying to get my head around power markets (European, but US responses are welcome) and particularly how traders decide which bets to take. My hypothesis is that these firms run models / forecasts about what is about to happen in the future, and then trade when they see variations between where they expect prices to go and where they are today. Assuming they're using models taking in some of the following data:
- Historic prices
- Prices for future settlement period
- Weather data
- Grid data
- Historic demand data
Would anyone be able to point me in the direction of the major types of models used at these firms and what input data they use. Interested in both long-term and short-term forecasting. Or just generally illuminate how this is done. Or also just call out if this is not at all how decisions are made on power desks.
Thanks in advance, and apologies if pretty basic (I tried searching but couldn't find a similar question)
Hi WannabeWattTrader, check out these threads:
More suggestions...
You're welcome.
Bump
Most short-term power trading is indeed very model driven and almost algorithmic. These desks usually build one massive model with the data you mentioned except historical prices. Normally marginal costs are calculated based on live prices of other commods and then the model computes where the auction will clear. Don't overestimate the complexity, often these models are simple linear regressions since they are more robust and easier to interpret. The difficulty of the problem lies in the breadth and data gathering process itself. Garbage in, garbage out.
More long-term trading is less model intensive since there are a lot more known and unknown unknowns. Usually more risk/reward style trading based on some fundamental idea.
That's super useful, thank you!
Two followups:
- Could you unpack the data difficulty? What makes it so hard? How do you solve it currently?
- On the linear regressions, how is that used in an algorithmic sense? I understand mathematically these can describe relationships between variables, but how do you use it to make automated trading decisions? (point me to some reference material if easier)
The day-ahead auction clears where demand meets supply so you start modeling demand and supply backwards with a bunch of different submodels all with varying complexity. Data is often bad but you wont know this unless you're working on a daily basis with this stuff.
The day-ahead auction clears where demand meets supply so you start modeling demand and supply backwards with a bunch of different submodels all with varying complexity. Data is often bad but you wont know this unless you're working on a daily basis with this stuff.
Never understand why people rush to model the whole situation in energy. Just becomes noise as there is so much one does not yet understand. Versus if focusing on just “load/demand forecast” to start you would get the mechanics of the market so much better.
It seems like power is the ideal commodity to work in with the world electrifying in many aspects. And unlike other commodities (especially ags), it's much harder to have a period of oversupply given how difficult it is to build capacity so volatility will be high going forward. Also, doesn't rely on having as much people skills as other commodities so as long as you are likeable enough and have the technical skills you're golden.
Anyone agree/disagree?
Don't over-complicate things like one pointed out above. Focusing on the load to start and there are tons of ways to analyze this (e.g. vendors' load forecasts error vs. ISO/RTO forecasts vs. actual). To take that one step further, apply some sort of probabilistic load model and translate that into power prices. Now you have your distribution of what load could be coming at, apply that step to the other variables that contribute to prices and dependent on which market you are covering, that can be somewhat different (wind in SPP/ERCOT is going to be much more important than wind in PJM for example)
Traders, regarding the processing the ''bad''/''garbage'' data from the various sources into ''good''/''clean'' usable data for the models:
1. How much time do you spend on this in a given week?
2. Is this one of the ''less interesting'' parts of the job? Something that must be done well but you'd rather you didn't have to? In my humble experience, data gathering and cleansing being less exciting than the model building and iteration.
3. Is this process that you would do as a trader, or typically outsourced to a data scientist/data engineer on the team, or even outsourced externally? Is there a reason why this should not be outsourced?
Really appreciate it! Trying to understand more about the day-to-day especially when it comes to data.
Congestion
Asperiores similique dolor rerum. Aut doloribus delectus neque qui ab nesciunt ea. Quam qui vero quia atque adipisci. Vel non nihil adipisci quae. Molestiae at libero blanditiis voluptatem voluptatem. Totam nesciunt quasi et sit.
Quod et dolor aut numquam. Repellat minima saepe corrupti. Quas nulla distinctio enim. Corrupti nisi ut quisquam porro voluptas.
Nemo qui rerum numquam molestiae. Voluptas eligendi numquam tenetur voluptatem. Ex nihil aspernatur corporis cum quidem voluptatem repudiandae saepe. Eligendi ipsam tenetur iste non. Eius consectetur eligendi ipsa maiores magnam veritatis.
See All Comments - 100% Free
WSO depends on everyone being able to pitch in when they know something. Unlock with your email and get bonus: 6 financial modeling lessons free ($199 value)
or Unlock with your social account...