Topic 9 – Mean Reversion Intuition in Commodities

1. What is mean reversion, in plain language?

Mean reversion = “pulled back toward a typical level”.

• If price is high, drift is downwards (all else equal).
• If price is low, drift is upwards.

Simple continuous-time cartoon (Ornstein–Uhlenbeck):

$$dX_t = \kappa (\theta - X_t)\,dt + \sigma\,dW_t$$

• $X_t$: state (e.g. log spot, short-rate, spread).
• $\theta$: long-run mean level.
• $\kappa > 0$: speed of mean reversion.
• $\sigma$: volatility.

If $X_t$ is above $\theta$ → drift negative; If below $\theta$ → drift positive.

2. Why commodities often behave mean-reverting

3. Why power is “mean reverting plus spikes”

Power is special:

• Electricity is hard/expensive to store directly.
• Demand must be balanced instantly, every hour.
• Small shocks (plant trips, weather surprises) can cause price spikes.

So spot power is more like:

• Fast mean reversion around a structural level,
• Plus occasional jumps/spikes and strong seasonality.

Many models use:

• Mean-reverting core (e.g. OU or mean-reverting log-price),
• Additive or multiplicative jump components.

Forward/average prices (month, quarter, year) look much smoother, but the intuition is still:

“Shocks to fundamentals eventually wash out; you don’t get unbounded exponential drift forever.”

4. Comparison vs GBM (equity-style)

GBM (equity):

$$\frac{dS_t}{S_t} = \mu\,dt + \sigma\,dW_t$$

• Drift is constant in relative terms.
• Price has no natural anchor; it can trend away indefinitely.
• Good-ish as a first approximation for equity indices.

Mean-reverting commodity model (very rough):

$$d(\ln S_t) = \kappa(\theta - \ln S_t)\,dt + \sigma\,dW_t$$

• Modelled state is log-price.
• If price gets very high vs $\theta$, drift points down, and vice versa.
• Long-run distribution of prices is stationary around $\theta$ (under a given measure).

Intuition to say aloud:

“Equity prices are claims on growth and can drift upwards indefinitely. Commodity spot and short-dated forwards are anchored by production costs, storage, and demand, so they revert toward a fundamental band. That’s why OU-type mean-reverting processes are natural building blocks in power & gas models.”

5. How this feeds into forwards and risk

Even if we model spot as mean-reverting, we usually trade forwards:

• Under risk-neutral measure, the forward is the risk-neutral expectation of spot.
• For mean-reverting spot, the forward tends to be a smoothed version of spot:
  • Far into the future, its level tends to the long-run mean (adjusted for risk premia).
  • Short-dated forwards track spot more closely.

Risk consequences:

• Short-dated risk: more influenced by current spot deviations and spikes.
• Long-dated risk: more influenced by changes in long-run mean $\theta$, not current noise.

Portfolio P&L is thus sensitive to:

• Shocks to current spot (short tenors),
• Re-estimation of long-run level and mean-reversion speed (long tenors).

6. Interview soundbites

A short way to talk about it:

“In commodities, fundamental feedbacks and storage economics pull prices back toward a cost band. That’s why we often model short-dated factors as mean-reverting diffusions, sometimes with jumps. Power spot is even more extreme: fast mean reversion with spikes and strong seasonality.

Compared to GBM in equities, mean-reverting models encode that prices don’t drift off to infinity on fundamentals alone; shocks decay over time, and long-dated forwards reflect the long-run mean more than current spot noise.”