Evaluating the Effectiveness of RGGI

class: center, middle, inverse, title-slide

.title[
# Evaluating the Effectiveness of RGGI
]
.subtitle[
## A Cap-and-Trade Analysis of CO₂ Emissions in the Eastern United States
]
.author[
### Hanyu Zhang & Catalina Toro
]
.institute[
### Columbia University | SIPA
]
.date[
### Spring 2026
]

---

# What is RGGI?

**The Regional Greenhouse Gas Initiative (RGGI)** is the first mandatory cap-and-trade program in the United States targeting CO₂ emissions from the power sector.

- **Scope:** A cooperative effort among **11 northeastern U.S. states** (CT, DE, MA, MD, ME, NH, NJ, NY, PA, RI, VT)

- **Coverage:** Applies to **regulated fossil fuel-fired power plants** with capacity **≥ 25 MW**   *(≥ 15 MW in New York)*.

- **Mechanism:** Participating states set a **regional cap on CO2 emissions** that declines over time. Regulated plants must hold **one allowance for each short ton of CO2 emitted**.

- **Auctions:** Allowances are distributed primarily through **quarterly auctions**. The proceeds are largely reinvested in **clean energy, energy efficiency, and bill assistance programs**.

- **Timeline:** Launched in **2009**; has undergone three Program Reviews (2013, 2017, 2025) to update the cap trajectory and program design

---

# The Identification Challenge

Simply comparing emissions before and after 2009 in RGGI states is **not enough**.

Emissions change for many reasons unrelated to RGGI:

- The **2008–2009 financial crisis** reduced electricity demand across the U.S.

- The **shale gas revolution** made natural gas cheap, displacing coal nationwide — including in non-RGGI states

- **Renewable energy growth** and efficiency improvements affected all states

---

# Research Question

### Main Question
**Did RGGI reduce CO₂ emissions from regulated power plants?**

### Next Question
**If yes, through what channel?**  
Did plants become **less carbon-intensive**, or did they simply **run less**?

<br>

This distinction is important: lower **total emissions** do not necessarily mean cleaner production. Emissions may fall because plants generate less electricity, or because they emit less CO₂ for each MWh produced.

---

# Data Sources

.pull-left[
### Emissions Data
- **Source:** EPA Clean Air Markets Program Data (CAMPD)
- **Unit:** Individual generating unit (boiler/turbine)
- **Period:** 2000–2025
- **Variables:** CO₂ emissions, gross load (MWh), heat input (mmBtu), fuel type, program code
]

.pull-right[
### Facility Attributes
- **Source:** EPA Facility Attributes (CAMPD)
- **Period:** 2000–2025
- **Variables:** Nameplate capacity (MWe), unit type, operating status
- **Key use:** Extract capacity to enforce coverage threshold filter
]

---

# Sample Construction

We aggregate from unit-year to **facility-year** level and apply five filters:

**1. Electric sector only**  
Remove non-electric units (e.g. Process Heaters and Cement Kilns)

**2. Capacity threshold — "apples to apples"**  
RGGI only covers plants **≥ 25 MW** (**≥ 15 MW in New York**). We apply this threshold to *all states*, so the control group contains only plants that *would have been covered* had their state joined RGGI

**3. Positive emissions**  
Drop plant-years with zero or missing CO₂

**4. Exclude leaker states**  
Remove states that may absorb shifted generation from RGGI states (**PA, OH**)

**5. Exclude non-continuous RGGI participants**  
Remove states that did not participate consistently over the study period (**NJ, VA**)

---

# RGGI Participation by State

|State | RGGI Plants |Participation               |
|:-----|:-----------:|:---------------------------|
|NY    |     86      |Since 2009                  |
|PA    |     58      |Since 2022                  |
|NJ    |     43      |2009–2011; rejoined in 2020 |
|MA    |     31      |Since 2009                  |
|VA    |     26      |2021–2023 only              |
|MD    |     21      |Since 2009                  |
|CT    |     19      |Since 2009                  |
|DE    |      9      |Since 2009                  |
|ME    |      6      |Since 2009                  |
|RI    |      6      |Since 2009                  |
|NH    |      5      |Since 2009                  |
|VT    |      2      |Since 2009                  |
---

# Difference-in-Differences Design

**Core idea:** Compare how emissions changed over time for RGGI plants *relative to* similar Non-RGGI plants. The difference in trends is attributed to the policy.

`$$\log(\text{CO}_2)_{it} = \alpha_i + \lambda_t + \beta \cdot \text{DID}_{it} + \varepsilon_{it}$$`

| Term | What it does |
|------|-------------|
| `$\alpha_i$` | **Plant fixed effects** — absorb time-invariant plant characteristics |
| `$\lambda_t$` | **Year fixed effects** — absorb economy-wide shocks affecting all plants |
| `$\text{DID}_{it}$` | **= 1** if plant is RGGI-covered **and** year ≥ 2009; **= 0** otherwise |
| `$\beta$` | **The causal effect of RGGI** |

- Standard errors **clustered at the facility level**
- Dependent variable: **log(CO₂)** — coefficient interpreted as % change
- Second specification adds **log(heat input)** to control for plant activity level

---

# Key Assumption: Parallel Trends

The DID design requires that, **absent RGGI**, treated and control plants would have followed similar emissions trends.

We begin with a simple visual check: comparing average emissions for **ever-RGGI** and **never-RGGI** plants over time.

---

# Parallel Trends: Event Study Evidence

We next test the parallel trends assumption more formally using an **event study** specification.

- The reference year is **2008** (`k = -1`)

- The weighted event study shows a gradual negative post-treatment pattern, suggesting that plants in RGGI states reduced emissions more relative to the control group after the policy began. However, the pre-treatment coefficients remain mostly positive, so the parallel trends assumption is still not fully convincing.

---

# Main DID Result

.center[
| | Gross-load-weighted TWFE |
|---|---:|
| **DID coefficient** | -0.250*** |
| **Std. Error** | (0.070) |
| **P-value** | 0.0004 |
| **Plant FE** | ✓ |
| **Year FE** | ✓ |
| **Weights** | Gross load |
]

<br>

### Key Finding

The weighted DID estimate is **negative and statistically significant**. This implies an estimated reduction of about **22%** in plant-level CO₂ emissions.

.small[Because the regression is weighted by gross load, this estimate gives more weight to larger generating plants.]
---

# Mechanism Results

.center[
| Outcome | DID coefficient | Std. Error | P-value | Within R² |
|---|---:|---:|---:|---:|
| **log(heat input)** | -0.229*** | (0.065) | < 0.001 | 0.0057 |
| **log(gross load)** | -0.225** | (0.070) | 0.0013 | 0.0046 |
| **log(CO2 intensity)** | -0.025 | (0.034) | 0.4513 | 0.0004 |
]

<br>

### Interpretation

- **Heat input** measures fuel use.  
  The coefficient of **-0.229** implies that treated plants used about **20.5% less fuel** after RGGI.
- **Gross load** measures electricity generation.  
  The coefficient of **-0.225** implies that treated plants generated about **20.1% less electricity** after RGGI.
- **CO₂ intensity** measures emissions per unit of output.  
  The coefficient is **small and statistically insignificant**, suggesting little evidence that plants became substantially cleaner per unit of electricity generated.
  
---
### Conclusion

The decline in **CO₂ emissions** appears to come mainly from:

- **lower fuel use**
- **lower electricity generation**

rather than from a significant improvement in **emissions efficiency**.

---
# Next Steps

**1. Check Parallel Trends**

Our event-study results suggest some pre-trend differences, so we will further test the DID identification strategy using placebo tests, alternative control groups, and more restricted samples.

**2. Emissions Leakage**

Did reductions in RGGI states simply shift emissions to non-RGGI states? We will examine whether Non-RGGI plants near RGGI borders increased output after 2009, using imports data from RGGI's monitoring reports.

**3. Heterogeneous Effects by Fuel Type**

Did coal plants respond differently than gas plants? Carbon pricing creates stronger incentives for higher-emission units — we expect larger effects for coal-heavy plants.

**4. Staggered Treatment**

Pennsylvania (2022) and Virginia (2021) joined late. We will use the **Callaway & Sant'Anna (2021)** staggered DID estimator to avoid bias from two-way fixed effects with heterogeneous treatment timing.

---
class: inverse

<h1 style="margin-bottom: 28px;">Thank You</h1>

<p style="font-size: 1.2em; margin: 0 0 14px 0;">Hanyu Zhang &amp; Catalina Toro</p>

<p style="font-size: 1em; margin: 0;">Columbia University | SIPA | Spring 2026</p>

</div>