📈 economics

Global Green Knowledge Flows: Analysis Based on Green Patent Network

Analyzing 6.65 million global green patents from 1990 to 2019, this study reveals that the inherent instability and shifting centrality of green patent networks—from end-of-pipe to preventive technologies—render standard prediction models ineffective, necessitating a newly developed dynamic rolling prediction approach to accurately forecast green innovation.

Original authors: Meiling Kang, Mengru Wang, Yechi Wu, Zhongkuang Zhao, Xiaoyan Zhou

Published 2026-07-06

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Meiling Kang, Mengru Wang, Yechi Wu, Zhongkuang Zhao, Xiaoyan Zhou

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: Tracking the "Brain" of Green Tech

Imagine the world's green technology efforts (like solar panels, water filters, and electric batteries) as a giant, living library. Every time someone invents a new green gadget, they write a "patent," which is like a new book in this library. When inventors create something new, they often look at old books (previous patents) for inspiration. This paper looks at 6.65 million of these "books" published between 1990 and 2019 to see how ideas are moving around.

The authors wanted to answer two main questions:

How is this green technology library different from a normal technology library?
Can we predict what the next big green invention will be?

The Big Discovery: Green Tech is a "Chameleon," Not a Rock

Most previous studies (like the famous work by Acemoglu) looked at general technology and found that the network of ideas was stable. They found that if you knew how ideas flowed in 1990, you could guess how they would flow in 2010 because the structure stayed the same. It was like a rock: solid and unchanging.

However, this paper found that Green Technology is not a rock; it is a chameleon.

The Shift: The "center of gravity" in green tech has moved dramatically.
- In the 1990s: The most important ideas were about cleaning up messes (end-of-pipe solutions). Think of it like focusing on scrubbing a dirty floor or filtering smoke coming out of a chimney. The "stars" of the network were water treatment and waste incineration.
- In the 2010s: The focus shifted to preventing the mess (preventive solutions). The "stars" changed to clean energy, batteries, and recycling. The network moved from "fixing the problem after it happens" to "stopping the problem before it starts."
The Result: Because the network keeps changing its shape and focus so quickly, the old maps (prediction models) that worked for stable industries don't work here. Trying to predict green tech using old methods is like trying to navigate a river that changes its course every week using a map from last year.

The Solution: A "Rolling Camera" Instead of a "Still Photo"

Since the green tech network is unstable, the authors realized they needed a new way to predict the future.

The Old Way (Static Model): Imagine taking a still photo of a moving car and trying to guess where it will be in 10 seconds. If the car turns a corner, your guess will be wrong. This is what traditional models did; they assumed the network was a still photo.
The New Way (Dynamic Rolling Model): The authors built a rolling camera (a video). Instead of looking at one fixed snapshot, their model looks at the network in moving time windows. It constantly updates its understanding of how ideas are flowing right now to predict what will happen next.

The Test:
They tested both methods.

The Old Method failed to predict the future accurately because it couldn't handle the changes.
The New Rolling Method was incredibly accurate. It successfully predicted over 90% of future green innovations by watching how the "flow" of ideas changed over time.

Key Takeaways from the Data

Some Fields are Superstars: Certain areas, like water treatment and electricity storage, are the "hubs" of the network. Almost everyone cites them. They are the "main characters" in the story of green tech.
Some Fields are Isolated: Other areas are like "silos." They don't talk to many other fields. They are very specialized and don't share ideas as much.
The Network is Getting Tighter: Over the 30 years, the different green fields started connecting more with each other. The "clustering" increased, meaning different types of green tech are starting to borrow ideas from one another more often.

Why This Matters (According to the Paper)

The authors suggest that because green technology changes so fast, we can't use "one-size-fits-all" rules to predict it.

For Policymakers: You can't just set a policy and forget it. You need a system that adapts as the technology evolves, just like the "rolling camera" model.
For Researchers: To understand green innovation, you have to look at the internal flow of ideas, not just outside factors like laws or money. The way ideas spill over from one field to another is the real engine driving change.

In short: Green technology is a fast-moving, shape-shifting river. The old maps don't work, but this new "rolling camera" method can finally help us see where the river is going next.

Technical Summary: Global Green Knowledge Flows: Analysis Based on Green Patent Network

1. Problem Statement

While extensive research exists on how external factors (regulations, policy instruments, market shocks) influence green innovation, there is a significant gap in understanding the internal mechanisms of knowledge flow, structure, and accumulation within green technology domains. Existing literature, such as Acemoglu et al. (2016), suggests that innovation networks are generally stable, allowing for prediction based on historical citation patterns. However, it remains unclear whether these stable network assumptions hold true for green technologies, which face unique interdisciplinary challenges and dual objectives (economic and environmental).

The core problem addressed is twofold:

Structural Instability: It is unknown how the green patent network differs from traditional patent networks in terms of stability and temporal evolution.
Predictive Failure: Standard innovation prediction models, which assume network consistency, may be ineffective for forecasting green innovation due to the potential time-varying nature of green knowledge flows.

2. Methodology

Data Sources and Construction

Dataset: The study utilizes 6.65 million global green invention patents filed between 1990 and 2019.
Selection Criteria: Utility model and design patents were excluded. Patents were screened using the "IPC Green Inventory" classification criteria developed by the International Patent Classification (IPC) Expert Committee.
Network Construction: A knowledge input-output network was constructed based on backward patent citations. The weight $w_{ij}$ represents the citation share of technical field $i$ citing field $j$ :
$w_{ij} = \frac{cite_{ij}}{\sum_k cite_{ik}}$
Where $cite_{ij}$ is the number of citations from field $i$ to field $j$ . This matrix $\Omega$ maps the knowledge spillover and flow between different green technology fields.
Classification Robustness: The study tested multiple classification granularities to ensure robustness:
- First 2 digits of IPC.
- First 4 digits of IPC.
- "IPC Green Inventory" categories (7 main categories).
- "IPC Green Inventory" sub-categories.
- Full IPC numbers.

Analytical Approach

Network Metric Analysis: The study calculated complex network indicators (e.g., degree centrality, network density, clustering coefficient, modularity, Gini coefficient) across three time intervals (1990–1999, 2000–2009, 2010–2019) to characterize the network's evolution.
Comparative Prediction Models: Two distinct forecasting approaches were tested using Ordinary Least Squares (OLS) regression:
- Traditional Static Lag Model: Based on Acemoglu et al. (2016), this model assumes a stable network. It calculates a fixed citation flow ($CiteFlow$) over a 10-year window and applies it as a static lag to predict future patent counts.
- Dynamic Rolling Prediction Model: Proposed by the authors to address network instability. This method treats the citation flow ($CiteFlow$) as a time-varying variable. It utilizes rolling time windows to dynamically adjust the citation flow coefficients, capturing the evolving nature of knowledge spillovers.

3. Key Contributions

Identification of Network Instability: The paper demonstrates that the green patent network is fundamentally different from traditional innovation networks. Unlike the stable networks described in prior literature, the green network exhibits pronounced spatio-temporal instability. Key nodes shift significantly over time, and the network structure transitions from end-of-pipe technologies (e.g., waste incineration) toward preventive solutions (e.g., clean energy, resource management).
Methodological Innovation: The authors propose a dynamic rolling prediction approach specifically calibrated for networks with evolving knowledge patterns. This method overcomes the limitations of static models that assume network consistency, providing a new tool for analyzing unstable, time-varying networks.
Endogenous Driver Analysis: Moving beyond the focus on exogenous drivers (policy, market), the study elucidates the internal driving mechanisms of green innovation by analyzing the dynamic role of patent networks and knowledge flows.

4. Results

Network Characteristics

Spatial Imbalance: The network shows uneven spatial distribution. Core nodes (e.g., water treatment $C02F$ , separation techniques $B01D$ , power supply $H02J$ ) dominate, while peripheral nodes (e.g., specific motor systems, fluid heaters) have limited connectivity.
Temporal Evolution:
- Connectivity: Network density increased significantly from 0.2023 (1990–1999) to 0.3969 (2010–2019), indicating more frequent knowledge exchange.
- Centrality: The network became less centralized (declining centrality index and Gini coefficient), suggesting a shift from a few dominant nodes to a more decentralized structure where multiple nodes share importance.
- Structural Shift: The top nodes shifted from traditional mechanical/combustion technologies ( $F01N, F23G$ ) in the 1990s to clean energy, storage, and recycling technologies ( $H02J, H01M, B09B$ ) in the 2010s.
Knowledge Source Trends: The proportion of green patents citing other green patents rose sharply (1990–1993), declined (1993–2006) as non-green fields became crucial, and then returned to growth (post-2006), indicating a transition from vertical specialization to horizontal expansion and back to specialization.

Prediction Performance

Static Model Failure: The traditional static lag prediction model failed to produce significant correlations between predicted and actual values when applied to the green patent network. The authors attribute this to the model's inability to account for the network's inherent structural instability.
Dynamic Model Success: The proposed dynamic rolling prediction model achieved high accuracy. Across four different classification criteria, the $R^2$ values exceeded 0.93, with one reaching 0.978. This indicates that the dynamic model can explain over 90% of future green innovations based on past decade knowledge flows, significantly outperforming static methods.

5. Significance and Claims

The paper claims that its findings offer a robust methodological framework for predicting future green innovations, which is critical for addressing global sustainability challenges.

For Policymakers: The research provides a tool to move from "reactive response" to "proactive shaping" in environmental governance. It suggests that policies should not be static; instead, incentives should be dynamically optimized to align with the evolving structure of green technology networks. It warns against "one-size-fits-all" subsidy approaches.
For Enterprises: The study highlights the importance of interdisciplinary cooperation and leveraging existing technological resources. It suggests that firms should focus on integrating technologies across shifting network centers (e.g., moving from waste treatment to energy storage).
For Researchers: The paper establishes that knowledge flows are a critical endogenous driver of green innovation. It provides a new perspective for understanding the generation of environmental technologies and offers a validated method for studying time-varying innovation networks.

The authors conclude that by revealing the dynamic essence of knowledge flows, their findings assist stakeholders in strategically allocating resources to foster green technology development, thereby contributing to the Sustainable Development Goals (SDGs), particularly regarding climate change and high-quality economic growth.

Note: The authors acknowledge limitations, including reliance on IPC classification for patent screening and the assumption that 10-year citation windows fully capture knowledge flow. They suggest future work could integrate machine learning (e.g., LLMs) for better patent identification and incorporate exogenous factors like policy shocks to further refine prediction accuracy.