Renewable Momentum and Forecast Decoupling in Power Markets

[WITH CODE] Building a Conditional Alpha Model with Wind and Load Velocity Signals

Hello and welcome back to another paid post!

Today we will take a look at how out-of-sample physical grid signals can expose predictable volatility corridors, and how an expanding multi-factor regression can harvest those imbalances before real-time delivery.

Let’s dive right in.

---

Recall

Let’s review our baseline model from our first look at the Texas power grid from two weeks ago. By simply selling power in the Day-Ahead Market and buying it back in Real-Time every single hour of the year, we uncovered a persistent, structural average spread of $0.93 per megawatt-hour. It was a respectable baseline, but trading every single hour of the year regardless of external signals is a heavy-handed tactic. In the real world, the friction of execution can quickly erode those theoretical gains. To evolve this baseline into something truly institutional, we have to look at how sophisticated desks trade the physical shape of the grid rather than just reading a static clock.

This transition from a naive clock to a selective, signal-driven framework is present in energy economics. In their seminal paper analyzing wholesale electricity markets, researchers Akshaya Jha and Frank Wolak discussed how financial virtual bids act as an essential arbitrage mechanism to align forward and spot prices. They examined the boundaries of market efficiency and concluded that convergence bidding profitability is structurally bound by transaction costs. Specifically, they noted in their conclusion that purely financial forward market trading can improve the operating efficiency of short-term commodity markets by optimizing thermal generation dispatch and reducing total variable production costs.

We can take this foundational thesis and extend it to modern intermittent grids, arguing that this spot price risk is driven by the real-time velocity of physical forecast errors. When the system operator’s day-ahead expectations decouple from physical reality, a violent real-time price correction occurs, and that is exactly where our new strategy hunts for alpha.

Jha, A., & Wolak, F. A. (2013). Testing for market efficiency with transactions costs: An application to convergence bidding in wholesale electricity markets. Stanford University. https://arefiles.ucdavis.edu/uploads/filer_public/2014/03/27/caiso_vb_draft_v8.pdf

---

Data Architecture

Before constructing a predictive engine, we must address the data required to blend financial prices with the physical state variables of the grid. Our asset universe consists of the same Day-Ahead Market (DAM) and Real-Time Market (RTM) as part one’s, using 2025’s data and HB_North prices.

We then will overlay system-wide load parameters and intermittent wind forecasting matrices. While the DAM closes at 10:00 AM, the physical inputs and resource planning schedules are locked in by grid operators much earlier in the morning. To reflect real-world desk execution and mitigate look-ahead bias, we discard any wind forecasts posted after 8:00 AM on the day prior to delivery. This way, we guarantee that our feature space matches the exact information pool available to a systematic participant before capital is committed.

---

Feature Engineering and Grid Physics

To capture conditional alpha, we must move away from static chronological filters and design features that directly map the physical stress states of the transmission network. The ERCOT power grid operates as a balancing act between fixed consumer demand and variable, intermittent renewable generation. Because electrical energy cannot be stored efficiently at a massive grid scale, any sudden divergence between the system operator’s day-ahead forecasts and the real-time physical dispatch forces the market to clear through violent price corrections. Our strategy targets this structural forecast decoupling by tracking the physical velocity of wind and load.