In 2DegreesInvesting/r2dii.physical.risk: Tools for Assessing Physical Climate Risk to Equity and Bond Portfolios

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

suppressPackageStartupMessages(library(dplyr))
library(r2dii.physical.risk)

This article documents the methodology used in this repository to estimate the risk a financial portfolio incurs as a result of physical hazards caused by climate change. The article assumes a basic understanding of PACTA.

Physical Risk

Climate change poses a financial risk from at least three different angles: Transition, Litigation and Physical. Transition risk is associated with the economic transition from high- to low-carbon technologies. If the transition is unmediated, it could cause a shock to the financial system. Litigation risk is the risk associated with potential lawsuits against companies that contribute significantly to climate change. Finally, physical risk, is associated with the real physical risk that climate hazards pose to assets. Here, we will focus on physical risks.

Physical risks can be classified as either acute or chronic. Acute risks refer to a frequency of isolated events, whereas chronic risks describe those risks that increase gradually and consistently.

• An example of an acute physical risk might be the risk of increased flood frequency on real estate.
• For chronic risks, you might imagine how the consistent increase in temperature, or precipitation, might affect agriculture.

As you can imagine, there are many different potential physical hazards that arise from climate change.

Canonical Risk Formulation

Physical risk can be expressed using the canonical risk formulation (below):

$$ Risk = Hazard * Exposure * Vulnerability $$ where:

• $Hazard$ quantifies the physical hazard itself (e.g. increase in river flood depth)
• $Exposure$ quantifies which assets will be affected by the $Hazard$, for example by number of assets or value of assets exposed (e.g which physical assets are located in areas with increased risk to river flood)
• $Vulnerability$ quantifies how much the $Hazard$ may damage the exposed assets (sometimes described as a "damage function")

Crucially, our model currently lacks "damage functions". This methodology currently spans only the $Hazard$ and $Exposure$ estimations.

Methodology

Let's go over the physical risk methodology presented in this repository.

Data

First, we will give a brief overview of the crucial datasets used in this methodology.

Input Portfolio

A typical "input portfolio" might look something like this:

portfolio <- tribble(
  ~isin, ~market_value, ~currency,
  "JP3868400007",       50000L,     "GBP",
  "FR0000571218",       28136L,     "GBP",
  "FR0011883966",      117291L,     "GBP"
)

portfolio

The input portfolio that is actually used in this physical risk assessment is already slightly pre-processed. The file used is the total_portfolio.rda file that can be found in the folder 30_Processed_Inputs of a typical PACTA project. This means that it has already been extended using financial data, and contains some columns that are not present in the above dataset. Crucially, it contains the following columns:

portfolio <- tribble(
  ~portfolio_name,    ~company_name, ~company_id, ~corporate_bond_ticker,          ~isin, ~holding_id, ~asset_type, ~security_mapped_sector, ~value_usd,
  "some portfolio", "a power company",          "C1",                  "ABC", "AB1234567890",        123L,    "Equity",                 "Power",     45000L
)

portfolio

Financial Data

Further financial data is used to map out the company ownership structure of each company in the portfolio. The goal is to be able to link each company that is present in the portfolio, all the way down to the individual physical assets that it might own and operate.

First, we need to understand which companies own the individual asset:

asset_level_owners <- tribble(
  ~company_id,        ~asset_id, ~ownership_share,
  "C2",    "A1",              25,
  "C3",    "A1",              25,
  "C4",    "A1",              50
)

asset_level_owners

In this case, we see that the power plant with ID "A1" is jointly owned by companies with IDs c("C2", "C3", "C4").

However, the only company in the portfolio (above) has the ID "C1"? We will see below that this company owns a stake of the above company.

To understand this, we need a dataset indicating the corporate ownership structure of each physical asset-owning company. This dataset looks something like this:

company_ownership_tree <- tribble(
  ~company_id, ~subsidiary_id, ~linking_stake, ~ownership_level,
          "C1",         "C2",             100,                2,
          "C2",         "C2",             100,                1,
          "C3",         "C3",             100,                1,
          "C4",         "C5",             100,                1
  )

company_ownership_tree

Physical Assets

The physical asset data used looks something like this:

ald <- tribble(
  ~asset_name, ~asset_location, ~latitude, ~longitude, ~sector,     ~technology, ~year, ~economic_unit, ~economic_output, ~asset_id,
  "Some Wind Farm",            "ES",       42L,        -3L, "Power", "RenewablesCap", 2020L,           "MW",             400L,      "A1",
  "Some Wind Farm",            "ES",       42L,        -3L, "Power", "RenewablesCap", 2021L,           "MW",             420L,      "A1",
  "Some Wind Farm",            "ES",       42L,        -3L, "Power", "RenewablesCap", 2022L,           "MW",             440L,      "A1"
)

ald

This data contains physical assets, including geographic latitude and longitude coordinates, for a variety of sectors, as well as production or capacity values for each asset.

Climate Hazard Scenarios

Climate Analytics

Climate data is "pre-processed", and harmonized with the physical asset data. For each combination of asset, scenario, model, period and hazard, there is an associated relative_change value. This value represents the estimated relative difference in the hazard (e.g. expected precipitation) in comparison to a reference period:

climate_data <- tibble::tribble(
  ~asset_name, ~scenario,     ~model, ~period,    ~hazard, ~relative_change, ~asset_id,
  "Some Wind Farm",   "rcp26", "Ensemble",   2100L, "prAdjust",            0.004,      "A1",
  "Some Wind Farm",   "rcp45", "Ensemble",   2100L, "prAdjust",             0.03,      "A1",
  "Some Wind Farm",   "rcp60", "Ensemble",   2100L, "prAdjust",             0.04,      "A1"
)

climate_data

Methodology

Now let's explain the methodology, by using the above datasets directly.

First, we expand each company in the portfolio by all of it's subsidiary companies. Companies with no matching data will be dropped.

out <- portfolio %>% 
  inner_join(company_ownership_tree, by = "company_id") %>% # join ownership
  inner_join(asset_level_owners, by = c(subsidiary_id = "company_id")) %>% # join asset owners
  inner_join(ald, by = "asset_id")

We then allocate an amount of economic output proportional to the ownership of each company as well as the ownership of the asset, to the portfolio:

out <- out %>% 
  mutate(
    portfolio_economic_value = (linking_stake/100) * (ownership_share/100) * economic_output
    )

  select(
    out,
    company_name, 
    value_usd, 
    asset_location, 
    latitude, 
    longitude, 
    sector, 
    technology, 
    year, 
    economic_unit, 
    portfolio_economic_value
    )

With the portfolio now linked to it's physical assets, and some proportional amount of production connected to each asset linked to the portfolio, we can begin to connect the climate hazards.

Since the hazards' geographic locations have already been harmonized to the asset locations, we simply match each asset to the portfolio by asset_id:

out <- out %>% 
  inner_join(climate_data, by = "asset_id") %>% 
    select(
    company_name, 
    value_usd, 
    asset_location, 
    latitude, 
    longitude, 
    sector, 
    technology, 
    year, 
    economic_unit, 
    portfolio_economic_value,
    scenario, 
    model, 
    period, 
    hazard,
    relative_change
    )

out <- out %>% 
  group_by(sector) %>% 
  mutate(
    portfolio_economic_value_share_sector = portfolio_economic_value/sum(portfolio_economic_value)
    ) %>% 
  ungroup()

We can now produce our output plots. These are simply bar charts that show how much of the production owned by the portfolio can be found in different risk quantiles. For the example above, consider we split the risk magnitude into a couple of chunks, $0 - 0.01$ relative change, $0.01 - 0.02$ and so on. We can display how much of the production owned by the portfolios is exposed to the low-risk quantiles ($0 - 0.01$ relative change) compared to how much is exposed to the high-risk quantiles. This gives us an overall estimate of how exposed the portfolio is to the particular hazard being studied.

provider <- "Fake Data Provider"
scenario <- "rcp26"
hazard <- unique(out$hazard)
model <- unique(out$model)
period <- unique(out$period)

r2dii.physical.risk:::plot_sector_absolute_portfolio_economic_value(
  out, 
  provider_sub = "Climate Data", 
  scenario_sub = scenario,
  hazard_sub = hazard,
  model_sub = model,
  period_sub = period
    )

2DegreesInvesting/r2dii.physical.risk documentation built on March 21, 2022, 2:03 a.m.