- wyatt8240
- Dec 8, 2025
- 3 min read
Many data providers have unique assets but struggle to articulate the investment case to institutional buyers.
That's the monetization gap we solve. Our collaboration with Kpler demonstrates the value.
The Challenge: Kpler tracks daily operational metrics for 900+ refineries globally. Hedge funds want earnings forecasts they can trade. But converting barrels-per-day into revenue predictions requires institutional quant expertise.
Most institutional investors wait weeks after quarter-end to see refinery earnings data. By then, the information advantage has evaporated. What if you could predict quarterly refining revenue within 0.5% accuracy—before the official announcement?
That's exactly what Exponential Technology achieved in partnership with Kpler, predicting Valero Energy's Q3 2025 refining revenue at $30.264 billion against the reported $30.415 billion. The 0.5% delta represents something rare in financial forecasting: actionable precision that translates directly into trading alpha.
Trade Refiners with Confidence Before They Report
This collaboration demonstrates a systematic methodology for forecasting public refinery company earnings using Kpler's daily operational refinery data combined with ensemble machine learning techniques developed by Exponential Technology.
Applied to three major U.S. refiners—Phillips 66 (PSX), Valero Energy (VLO), and PBF Energy (PBF)—the framework achieves mean absolute percentage errors (MAPE) of 3.13%, 1.86%, and 3.15% for quarterly production forecasts, respectively, and 3.10%, 4.30%, and 5.80% for quarterly refining revenue forecasts.
Key Takeaways
Low-error KPI forecasts across PSX, VLO, and PBF using daily refinery data and ensemble ML
Refining-revenue forecasts in low single-digit MAPE, powered by a unique "crack-ratio" method
Forward-validated trade example tied directly to pre-announcement model
Cumulative back test performance across multiple quarters and names
The Information Asymmetry Window
Kpler tracks daily operational metrics for 900+ refineries globally—throughput volumes, production rates, utilization, planned and unplanned downtime. This data updates at 8 AM UTC daily. Meanwhile, the same companies only disclose financial results quarterly via SEC filings, typically 1-4 weeks after quarter-end. That structural lag creates a predictable window where operational performance is known but earnings remain opaque.
Our research team converted Kpler's daily barrel-per-day metrics into revenue forecasts using a proprietary "crack ratio" framework that normalizes refining margins against WTI crude prices, creating a stationary, mean-reverting time series that's substantially easier to forecast than volatile absolute spreads.
The Q3 2025 Forward Validation
The real test came with Q3 2025 earnings, where we generated forecasts before announcements:
For Valero, our model predicted refinery throughput of 3,076 Kb/d versus reported 3,087 Kb/d (0.36% error) and refining revenue of $30.264B versus reported $30.415B (0.50% error). The forecast suggested a 9.5% upside to consensus estimates, triggering a long position signal that generated a 6.9% overnight return on the earnings announcement.
Historical backtesting across 30 trades (10 quarters × 3 stocks) produced 101% cumulative returns with a 70% win rate—substantially outperforming naive long-only (-46%) and short-only (+84%) benchmarks, with an estimated 40% average annual rate of return.
Exponential Technology Professional Services as Competitive Advantage
This collaboration exemplifies what our professional services practice delivers: we bridge the gap between raw data assets and institutional revenue. Many data providers possess unique information but struggle to articulate the precise investment case to sophisticated buyers. That's where quantitative validation creates commercial value.
For Kpler, we provided:
Novel Dataset Productization: Signal validation and conversion of daily operational metrics into predictive earnings models with institutional-grade performance metrics
Institutional-Grade Strategy Creation: Rigorous backtesting with Sharpe ratios, win rates, and Kelly Edge calculations that prove ROI to portfolio managers and quants
High-Impact Marketing Collateral: The comprehensive 16-page white paper with compelling visualizations and quantitative researcher credibility
The methodology is highly generalizable—applicable across Kpler's 900+ global refinery coverage and extendable to other commodity sectors where operational data leads financial disclosure.
From Data Entropy to Investment Intelligence
Alternative data only matters if it generates alpha. What institutional investors pay for isn't data access—it's the confidence that the data will predict something they can trade profitably. That requires the intersection of domain expertise, quantitative rigor, and enterprise-grade delivery infrastructure.
The full methodology, including mathematical derivations of the crack ratio framework, ensemble machine learning architecture, walk-forward validation protocols, and detailed performance metrics across multiple quarters, is available in our joint white paper with Kpler.
Comments