Hello,I'm trying to determine the mean reversion rates for energy futures. How far back do I go and pick the data to estimate the long run mean price (in $) and the mean reversion rate (in days) i.e. Do i use daily data for 6 months, 1 year, 2 years etc.?I'm using the analysis to develop some trading strategies with holding periods of anywhere between 1 week to a maximum of 6 months. Thanks.

I don't think data will be very useful in this regard. To get a useful parameter estimate for the type of trading you suggest, you would need many years of data. But I don't think energy mean reversion over that period is stable. If you take six months of data, you won't get a good fit. If you take 20 years of data, you'll be factoring in irrelevant economic environments.

I think you can refer to Les Clewlow 's book: Energy Derivativesricing and Risk Management. In his book ,he cite 0.003846 as timestep.

Thank you for your replies.I have used the Clewlow Energy Derivatives book as a reference and the 0.003846 time step implies a 1 year's worth of data. Aaron: I completely agree that 20 years of data will not adequately factor in current economic conditions. And yes the mean reversion rate is unstable over various time frames. For example, the NYMEX natural gas continuous futures contract, based on 1 year of the most recent data has a mean of $5.73 and reversion rates between 2 to 7 trading days or 3 to 10 actual days. As someone who follows energy futures explicitly, I can tell you that these reversion rates are not realistic. If you choose data for the past 2.5 years (from Jan'02) the mean price drops to $5.11 and reversion rates are between 4 and 69 trading days. I'm trying to think along the lines of statistical arb and something similar to pairs trading. But I guess stable mean reversion is required to be able to successfully implement that style and one is trading the spread, not the absolute price. Any suggestions? Intuitively, I'd like to think that for a 6 month holding period one should use 6 months of data.I guess the question to ask is: What is deemed long-term when it comes to energy prices? For a speculator, 6-months are long enough.And how practical is the entire concept of mean-reversion and will it apply to changing market fudamentals?

QuoteOriginally posted by: tmoiThank you for your replies.I have used the Clewlow Energy Derivatives book as a reference and the 0.003846 time step implies a 1 year's worth of data. Aaron: I completely agree that 20 years of data will not adequately factor in current economic conditions. And yes the mean reversion rate is unstable over various time frames. For example, the NYMEX natural gas continuous futures contract, based on 1 year of the most recent data has a mean of $5.73 and reversion rates between 2 to 7 trading days or 3 to 10 actual days. As someone who follows energy futures explicitly, I can tell you that these reversion rates are not realistic. If you choose data for the past 2.5 years (from Jan'02) the mean price drops to $5.11 and reversion rates are between 4 and 69 trading days. I'm trying to think along the lines of statistical arb and something similar to pairs trading. But I guess stable mean reversion is required to be able to successfully implement that style and one is trading the spread, not the absolute price. Any suggestions? Intuitively, I'd like to think that for a 6 month holding period one should use 6 months of data.I guess the question to ask is: What is deemed long-term when it comes to energy prices? For a speculator, 6-months are long enough.And how practical is the entire concept of mean-reversion and will it apply to changing market fudamentals?Perhaps you could start by choosing several time windows and get a qualitative feel as to the variabilility of mean reversion. Of course, the variability may be large, but you may find that it is small enough that it might fit your purposes - just a thought.

QuoteOriginally posted by: tmoiThank you for your replies.I have used the Clewlow Energy Derivatives book as a reference and the 0.003846 time step implies a 1 year's worth of data. Aaron: I completely agree that 20 years of data will not adequately factor in current economic conditions. And yes the mean reversion rate is unstable over various time frames. For example, the NYMEX natural gas continuous futures contract, based on 1 year of the most recent data has a mean of $5.73 and reversion rates between 2 to 7 trading days or 3 to 10 actual days. As someone who follows energy futures explicitly, I can tell you that these reversion rates are not realistic. If you choose data for the past 2.5 years (from Jan'02) the mean price drops to $5.11 and reversion rates are between 4 and 69 trading days. I'm trying to think along the lines of statistical arb and something similar to pairs trading. But I guess stable mean reversion is required to be able to successfully implement that style and one is trading the spread, not the absolute price. Any suggestions? Intuitively, I'd like to think that for a 6 month holding period one should use 6 months of data.I guess the question to ask is: What is deemed long-term when it comes to energy prices? For a speculator, 6-months are long enough.And how practical is the entire concept of mean-reversion and will it apply to changing market fudamentals?If you have 6-month holding period, you will need in general many years worth of data to get a sufficiently large set of non-overlapping samples. You can perform your estimates with overlapping data. There will be no bias, only the estimate will be very inefficient. Overall, as Aaron pointed out, it's not gonna work too well. It's largely a waste of time.