JPMorgan uses machine learning to forecast China's wind power generation, suggesting Longyuan's March data may face near-term pressure

Overview

This report focuses on China's wind power operators, especially China Longyuan Power. JPMorgan proposes using machine learning combined with wind-speed data from about 400 meteorological stations across the country, historical generation, installed capacity, and curtailment rates to forecast monthly wind power generation in advance, and to use that as a short-term trading reference before official generation data are released. The core conclusion is that Longyuan's wind power generation growth in March 2026 may lag installed-capacity growth, creating a mildly negative short-term stock reaction, while the medium- to long-term outlook remains supported by China's decarbonization, capacity expansion during the 14th Five-Year Plan, and a higher share of projects without subsidy arrears.

Core views

First, wind power operators' long-term generation usually tracks resource conditions and installed capacity, but monthly wind-speed swings can cause year-on-year generation growth to fluctuate sharply and affect short-term stock prices. Second, FY25 generation growth decoupled from installed-capacity growth, with generation growth at about 4%, below the roughly 16% installed-capacity growth guidance, due to higher curtailment rates and weaker wind speeds. Third, the model expects Longyuan's March 2026 wind power generation to be about 6 billion kWh, down about 8% year on year, below the roughly 6% organic installed-capacity growth, so the March generation release may bring a negative stock reaction. Fourth, despite the short-term pressure, the report still maintains Longyuan's Overweight rating, citing valuation, lower leverage, and a better regional capacity mix.

Analysis framework

The report first builds a monthly wind power generation forecast model, then compares the forecast with historical installed-capacity growth to identify potential 'abnormal' generation events. Inputs include wind speed from about 400 meteorological stations nationwide, historical generation data, installed capacity, and curtailment rates. A machine learning model is used for Longyuan's domestic portfolio; a simple time-series model is used for the roughly 1.5% of the portfolio in South Africa and Canada. The report also backtests a Longyuan trading strategy based on a 5% deviation threshold using 2019-2025 data.

Methodology notes

• Machine learning forecastMonthly wind power generation forecast model

  Forecast monthly generation using weather and operating data

  The model uses historical wind speed, historical generation, installed capacity, and curtailment rates to forecast Longyuan's monthly wind power generation and form a trading reference before monthly official data are released.

• Model error evaluationMAPE

  Mean absolute percentage error

  The report says the domestic portfolio model achieved an overall MAPE of about 5% on training and testing data from January 2014 to March 2026, which can be interpreted as an average accuracy of about 95%.

• Trading signal5% deviation threshold strategy

  Difference between forecast generation growth and historical installed-capacity growth

  If forecast generation growth exceeds historical installed-capacity growth by more than 5%, the sample strategy goes long; if it falls below by more than 5%, the sample strategy goes short. The report emphasizes that the backtest only targets the data component, and other market factors may still dominate the share price.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• China Longyuan Power / Longyuan (0916.HK)

  Core covered company and model example security

  Strengths

  One of China's largest and more mature wind farm operators; as of end-2024, consolidated wind installed capacity reached 30.4 GW; benefits from China's carbon-neutral development, capacity growth, and a higher share of projects without subsidy arrears; FY24 leverage was below DT RE.

  Weaknesses

  March 2026 wind power generation is forecast to decline by about 8% year on year, below capacity growth; generation is materially affected by monthly wind speeds, curtailment rates, and maintenance.

  Comparison

  The report believes Longyuan should deserve a higher target valuation multiple than DT RE because Longyuan's FY24 leverage of about 163% is below DT RE's level of over 300%, and its regional capacity mix is better.

  Risks

  Utilization hours below expectations, power prices below expectations, financing costs above expectations, and sudden capacity changes from M&A, unannounced maintenance at wind farms or regional grids, or a sharp rise in curtailment rates that the model does not capture.

Key data

• Longyuan March 2026 wind power generation forecastabout 6 billion kWhA 7.5% discount factor has been added to reflect the impact of changes in the curtailment-rate definition on the portion not covered by the model.
• Longyuan March 2026 generation YoY forecastabout -8%Below the roughly 6% organic installed-capacity growth.
• Average model accuracyabout 95%Corresponding to an overall MAPE of about 5% for the domestic portfolio.
• Trading strategy backtest59 trades, with an average absolute annualized return of about 11.0%Based on Longyuan forecasts from 2019 to 2025, with a 5% deviation between forecast generation growth and historical installed-capacity growth as the threshold.
• Longyuan domestic and overseas wind portfolioAbout 1.5% of the overseas portfolio is in South Africa and CanadaThe overseas portion uses a simple time-series model, which the report believes has a limited impact on overall accuracy.
• High-wind provincesYunnan +3.3%, Hainan +2.7%, Fujian -1.3%As of end-2025, Longyuan had installed wind capacity of 1,440 MW, 99 MW, and 1,053 MW in Yunnan, Hainan, and Fujian, respectively.
• Low-wind provincesTianjin -37.4%, Liaoning -28.9%, Hebei -26.6%As of end-2025, Longyuan had installed wind capacity of 582 MW, 1,640 MW, and 1,851 MW in Tianjin, Liaoning, and Hebei, respectively.
• Valuation basisDec-26 target price HK$8.6, target P/BV of 0.85xThe target multiple is about 0.9 standard deviations below the historical mean of 1.3x and reflects FY25E book value and impairment risk from retrofitting old units.

Impact & implications

The investment implications are two-tiered: in the short term, if the forecast of about -8% year-on-year growth in March generation is realized and clearly lags installed-capacity growth, Longyuan may face a negative stock reaction when the data are released; in the medium to long term, JPMorgan still sees value in China's wind power operators, supported by a higher share of projects without subsidy arrears, China's decarbonization policies, and accelerated capacity additions during the 14th Five-Year Plan. For investors, monthly wind-speed and generation forecasts can serve as an event-driven trading tool, but they cannot replace fundamental judgments on power prices, curtailment, financing costs, and policy discussions.

Risks

• Wind utilization hours below expectations.
• Power prices below expectations.
• Financing costs above expectations.
• The model may fail to fully capture sudden capacity changes caused by M&A.
• Unannounced maintenance at wind farms or regional grid issues.
• Power market disruptions that lead to a sharp rise in curtailment rates.
• Policy or market discussions may overwhelm the stock impact of generation data, such as the policy debate on FY21 green power trading that once dominated Longyuan's share price.
• Potential conflict-of-interest disclosures: J.P. Morgan may have business relationships, holdings, market-making activity, or client relationships with the covered company.

What to watch

• Longyuan's official release of March 2026 monthly generation data and the stock's reaction.
• Whether the deviation between forecast generation growth and historical installed-capacity growth exceeds the 5% threshold.
• Whether changes in the curtailment-rate definition and actual curtailment levels continue to affect model error.
• The drag on Longyuan's portfolio generation from low-wind regions such as Tianjin, Liaoning, and Hebei.
• The share of projects without subsidy arrears, grid-parity tariffs, financing costs, and impairment risk from retrofitting old units.
• The pace of China's decarbonization policies and wind capacity additions during the 14th Five-Year Plan.