Skip to main content
Skip table of contents

4.4.4 Robustness of the climate risk models

Robustness of the Regression Models

The table below presents the results of the regression analyses performed for both corporate and project companies. The regression coefficients are all highly significant. The R2 for both company types are relatively small (as expected when dealing with growth rates). Residuals are distributed close to normal, around 0, with very low autocorrelation levels. The results demonstrate a high level of robustness of our models.

Table: Regression results for corporate and project companies

 

Corporate companies

Project companies

Regressor

Total assets

Revenues

Total assets

Revenues

Total assets L1

0.078***

(0.016)

 

0.043***

(0.013)

 

GDP L1

0.344***

(0.076)

 

0.163***

(0.046)

 

Inflation L1

1.274***

(0.157)

 

0.631***

(0.087)

 

Percent Lifetime

 

 

-0.038***

(0.003)

 

Total asset

 

0.236***

(0.024)

 

0.243***

(0.018)

Constant

-0.006***

(0.002)

 

-0.006***

(0.002)

 

Observations

3,486

1,107

5,195

1,871

Adjusted R2

0.032

0.077

0.042

0.085

Note: L1 denotes the first lag (i.e., value in the previous year) of a variable. *p < .1; **p < .05; ***p < .01

Robustness of the Asset Pricing Model

For the 250+ transactions that correspond to companies tracked in the EDHECinfra universe and for which observed secondary market prices are also available, we compare observed and model-predicted valuations directly. The figure below shows a comparison between model-predicted internal rate of return (IRR; left panel), risk premia (central panel) and ratios of a company’s enterprise value to its earnings before interest, depreciation, taxes, and amortisation (EV/EBIDTA; right panel) with actual values for a test dataset of 250+ observed transactions that occurred between 2000 and 2020.

rpt_vs_est_irr_scatter.png

rpt_vs_est_equity_premia_scatter.png
ev_ebitda_scatter.png

Comparison between model-predicted IRR, risk premia, and EV/EBIDTA ratios with actual values for the test dataset of 250+ observed transactions between 2000 and 2020.

The closeness of the dots around the 45-degree lines and the high values of R2 support the accuracy of the model-predicted prices. The prediction error is typically within 5% of the observed prices (see tables below).

Table: Quantiles of model errors

10% Quantile

25% Quantile

Median

Mean

75% Quantile

90% Quantile

-5%

-1.95%

-0.22%

-0.55%

1.64%

3.85%

Table: Estimated vs reported valuation ratios and model’s goodness-of-fit

Ratio

Reported mean

Estimated mean

Reported median

Estimated median

R2

RMSE

EV/EBITDA

15.54

15.34

12.98

12.61

0.97

2.27

P/Book

2.37

2.28

1.65

1.59

0.87

0.9

P/Sales

3.35

3.21

2.52

2.32

0.85

1.43

Note: RMSE = root mean square error

JavaScript errors detected

Please note, these errors can depend on your browser setup.

If this problem persists, please contact our support.