TCFD climate stress test (2D)

A board of ten sector plots: set warming, carbon price, policy delay and pass-through, then read each sector's loss and the portfolio Value-at-Risk. Click a tile to focus on it.

Switch to 3D

Presets

Playback

Climate

Policy and companies

Stress test settings

Layers

Model check

The science

What a TCFD-style stress test does

The Task Force on Climate-related Financial Disclosures asks companies and investors to test their balance sheets against climate scenarios. Two channels matter: physical risk (floods, heat, drought damaging assets) and transition risk (carbon prices, new rules and technology changes repricing high-emission assets). This model reprices ten sectors worth $1,380 M in a stylised portfolio under a scenario you choose. All sector parameters are illustrative teaching numbers, not data for any real company, and the result is not a forecast.

The equations

Temperature path: T(t) rises linearly from 1.3 C in 2026 to your 2100 value. Physical loss of sector i: Lp = phii a (average discounted excess of T(t)2 over 1.32, from the stress date to 2100). Here a is the damage-function strength (output lost per degree squared; the DICE model uses about 0.27 %, later work about 0.75 % or more) and phii is how exposed the sector is compared with the economy average.

Carbon price: $25/t today, starting to rise after your policy delay and reaching your chosen value in 2040, flat afterwards. Transition loss: Lt = kappai (P40 - 25)/100 S (1 - rho pti), where kappai is the share of cash flow a $100/t price would remove, S the discounted share of the ramp, and rho pti the part of the cost passed on to customers. Fossil sectors also lose a stranded-asset share wi = si dP / (dP + 150).

Value: Vi = V0 (1 - Lp)(1 - Lt)(1 - wi). The two risks compound, so the portfolio loss is physical + transition minus an overlap.

Value-at-Risk

The central path is one point; reality is uncertain. 2,000 seeded Monte Carlo draws vary warming (spread 0.4 C), the damage coefficient (lognormal, spread 0.5) and the carbon price (lognormal, spread 0.4) together with the uncertainty slider. VaR95 is the loss that 1 draw in 20 exceeds; expected shortfall is the average of those worst draws. The same random numbers are reused when you move a slider, so the readouts change smoothly.

Honest simplifications

No adaptation spending, insurance pricing feedback, macro feedback, correlation between the three uncertainties, or cash flows beyond 2100. The temperature path is linear rather than emissions-driven. Sector figures are rounded stylised values; the scenario names echo common scenario families but contain no official scenario data.

Model checks (run live)

A closed form for the discounted average against a year-by-year sum; each risk alone reproduces its share of the combined loss; no shock gives zero loss; the Monte Carlo quantile of a lognormal against its analytic value; and monotonic response to warming, carbon price and pass-through.

Value lost by sector: physical vs carbon cost and stranding
Distribution of portfolio loss over 2,000 Monte Carlo draws
Warming and carbon-price paths to 2100
Portfolio loss against the stress date, for three scenarios
Made with 🟠 Blender