Open Thinkering


House purchases, climate change, and AI

Just over nine years ago, Team Belshaw needed to find somewhere to live. I’d been told by my employer at the time that, for various reasons, our dream of moving to Gozo couldn’t go ahead. But we still wanted to sell our house. So we moved from a four-bedroom detached house into a two-bedroom terrace in Morpeth.

It all worked out well. We ended up buying the house and converting the loft. I had a home office as the garage, separate to the house, was already converted for that purpose. Our neighbours have proved been amazing. Being so close to the centre of town and their schools it’s been a lovely place to raise our kids.

However, it’s never been our long-term plan to stay put; we’ve always been looking for somewhere else. The pandemic and then our son’s GCSE exams put off the sale of our house, but this year we finally put it on the market. Two weeks later we accepted an offer. On the same day, we had an offer accepted on a house which backs onto a river.

Black and white cropped version of a photo by Flickr user johndal showing Morpeth floods in 2008
Not this house! Just an illustrative photo of the 2008 floods.

Now, anyone who knows Morpeth knows that it flooded quite spectacularly in 2008 and then again in 2012. The Environment Agency has spent a lot of money on various flood defences including an underground reservoir and even the building of reasonably-attractive walls in the gardens of potentially-affected properties. In fact, they did such a good job that the risk of flooding from the river is negligible for the foreseeable future.

Today, though, we pulled out of the purchase of the riverside house. Why? We commissioned a more detailed flood report after someone had set alarm bells ringing about the impact of climate change. To cut a long story short, the property is at high risk of surface water flooding, and (in our opinion) the Environment Agency’s river flood defences potentially exacerbate that risk.

The reason I wanted to note this, other than it being a significant event in my life, is to share what I’ve learned about cumulative risk. I also wanted to note how helpful ChatGPT 4 was at summarising and making calculations based on the report. It was quite technical in places and, while data is good, we needed to extract meaning.

At the end of the day, humans should make decisions. But having a very quick and technically-proficient assistant to pull out and interpret data is extraordinarily useful.

I used the AskYourPDF plugin for ChatGPT, beginning by asking it specifically to explain “in the simplest language possible” the risk from surface water flooding. I then asked if we were planning to live there for 20 years whether it would be a sensible purchase. While it wouldn’t make a specific recommendation, its responses erred on the side of caution.

Going further, I asked about the impact of climate change on the property, to which it responded about things we’d already considered (increased rainfall, urbanisation), and things we hadn’t (sea level rise, decreased snowfall). This gave us further pause for thought.

The next day, already thinking that this wouldn’t be a wise investment, I asked ChatGPT to “come up with a range of 5 personas and explain how they would approach a purchasing decision for this property.” I specifically asked it to include the notion of “risk appetite” as part of this. The results were interesting. I followed up by enquiring “If these personas talked to one another and had to make one decision based on their discussion, what would be that decision?” Somewhat predictably, the personas consensus view was not to go ahead with the purchase.

The next step was to use it to calculate cumulative risk, which we’d realised hadn’t been specified in the report. It’s been a while since I did this kind of Maths, and doing it manually is somewhat confusing to my brain. So I asked ChatGPT:

Thank you for providing access to the document again. Based on the extracted information:

- The property is shown to be located within an area of **High surface water flood risk**.
- The definitions for the risk levels are as follows:
  - **High risk**: annual chance of flooding of greater than 3.3%.
  - **Medium risk**: annual chance of flooding of between 1% and 3.3%.
  - **Low risk**: annual chance of flooding of between 0.1% and 1%.
  - **Very Low risk**: annual chance of flooding of less than 0.1%.

Given that the property is in the high-risk category, the upper bound annual flood risk for surface water flooding is 3.3%. However, you've also asked for calculations for 5%, 6%, and 7%. We've already provided the 20-year cumulative risk for 5%, so let's calculate for 6% and 7%:

For 6%:
\[ \text{Risk over 20 years} = 1 - (0.94)^{20} \]
\[ \text{Risk over 20 years} \approx 74.0% \]

For 7%:
\[ \text{Risk over 20 years} = 1 - (0.93)^{20} \]
\[ \text{Risk over 20 years} \approx 80.7% \]

So, to summarize:
- For a 3.3% annual risk, the 20-year cumulative risk is approximately 49.4%.
- For a 5% annual risk, the 20-year cumulative risk is approximately 64.2%.
- For a 6% annual risk, the 20-year cumulative risk is approximately 74.0%.
- For a 7% annual risk, the 20-year cumulative risk is approximately 80.7%.

So a probability greater than a coin toss that, based on the current situation, the garden would be flooded. And this wasn’t taking into account other factors that weren’t covered by the data in the report.

The final step was to think about the risk of surface water flooding being so bad that it potentially enters the house. This was mentioned in the report as the ‘low-risk scenario’ of between 0.1% and 1% that up to 0.6m of water would be adjacent to the dwelling

So if we stayed there for 20 years, what's the cumulative risk from the annual low risk scenario?

Given the range for the low-risk scenario is between 0.1% and 1%, let's calculate the cumulative risk over 20 years for both the lower and upper bounds of this range.

Using the formula:
\[ \text{Risk over period} = 1 - (1 - \text{annual risk})^{n} \]

1. **For the lower bound of 0.1% (0.001)**:
\[ \text{Risk over 20 years} = 1 - (0.999)^{20} \]
\[ \text{Risk over 20 years} \approx 2.0% \]

2. **For the upper bound of 1% (0.01)**:
\[ \text{Risk over 20 years} = 1 - (0.99)^{20} \]
\[ \text{Risk over 20 years} \approx 18.2% \]

So, for the low-risk scenario:
- Over a 20-year period, there's approximately a 2.0% cumulative risk at the lower bound (0.1% annual risk) and an 18.2% cumulative risk at the upper bound (1% annual risk) of the property experiencing a flooding event.

So, even without the uncertainty of climate change, if we take the middle of the upper and lower bounds of cumulative risk, there’s a 1 in 10 chance that there will be up to 60cm of flood water adjacent to the property? No thanks.

Having pulled out of the purchase, this morning I’ve been to register our details with a bunch of estate agents for rentals. The chances are we’ll have to bide our time to get the exact house we want. That will be annoying, but not too much of a hardship. Especially as the next house we buy we’ll probably be staying in for a while.

Photo modified from an original CC BY SA johndal

4 thoughts on “House purchases, climate change, and AI

  1. Wow this is fascinating! Our house report was so generic and covering the assessors behind. It wasn’t really worth the paper it was printed on. Not sure how chatGPT might fair if I gave it that as a PDF beyond the shrugging emoji!

    Also dont (yet) have ice cream maker but where is recipe details please

  2. Fascinating stuff Doug, really interesting to see how you used this tool. Some great techniques. Sorry the house didn’t work out, but right from the start of this post I was concerned – I’m sure you made the right choice. More broadly, I do wonder how many people are likely to start to see similar properties as unsafe investments and their value start to tank. As owner of a victorian terrace I wonder the same about properties with insulation setups based on century old technology and climate expectations.

Leave a Reply

Your email address will not be published. Required fields are marked *