JERK Report #5 Graph Paper

In 1986, I sat on a bed watching my grandfather plot data on graph paper.

He was a physicist. He was mapping historical temperature records, trying to determine whether climate change was real — or whether it was just normal variation if you looked at a wider time period.

At the time it was interesting, but not unusual for him. He was always plotting something.

It was one of the last conversations I had with him before he died. So it stayed with me.

This week, it came back.

Today

I'm writing this from Maryland on March 16, 2026.

The National Weather Service has issued a Level 4 out of 5 severe weather risk — from Maryland to South Carolina. Tornado watch for 17 counties. Wind gusts forecast at 70 to 80 mph. Potential for significant tornadoes. Governor Moore has declared a State of Preparedness. Schools closed early.

This is the second tornado watch in Maryland in five days. The first was March 11.

Maryland's traditional tornado window is May through July. It's mid-March.

With the threat of 2-inch hail hitting my neighborhood today, my grandfather's question came back to me.

So let's look at the data. Through the lens of jerk.

Position: Everything looks fine

If you check the total number of hailstorms per year in the U.S., the count is basically flat.

Total tornado count since 1970? Also flat.

If you only read position — the current snapshot — you'd say nothing has changed.

Most people stop here.

Velocity: Things are moving

Billion-dollar weather disasters averaged 3 per year in the 1980s. Over the last decade, that number is 19 per year.

The average time between billion-dollar disasters was 82 days in the 1980s. By the last decade, it was 19 days. In 2023 and 2024, it was every 12 days.

All figures adjusted to 2024 dollars.

Things are moving.

Acceleration: The rate of change is increasing

Average annual disaster costs: $22 billion per year in the 1980s. $99 billion in the 2010s. $149 billion in 2020–2024.

That's not steady growth. The 2020s jumped 50% over the 2010s in just five years.

The rate of change is itself accelerating.

Jerk: What the insurance industry already knows

Here's where it gets interesting.

The total count of storms isn't changing much. But the composition is shifting.

Hailstorms producing 2-inch stones: 4% of all hailstorms in 2000. 12% in 2024. Both 2023 and 2024 set records.

High wind events: 16,123 incidents in 2000. 25,371 in 2024 — a steady 1.83% compound annual growth rate over 25 years.

Tornado activity is shifting east — from the Plains into the Southeast and Mid-Atlantic — and arriving earlier in the year. March 2025 broke records with 299 reported tornadoes, nearly four times the 30-year average. The first EF5 tornado in over 12 years hit in June 2025. Severe weather accounted for a record 21 billion-dollar disasters in 2025 alone.

The pattern of the pattern is changing.

And the insurance industry — whose entire business is pricing risk — has noticed.

State Farm stopped writing new homeowner policies in California in 2023. Then non-renewed 72,000 policies. Then projected dropping 1 million by 2028.

In Louisiana, 30 to 40% of real estate deals now collapse because of insurance quotes.

7.4% of U.S. homeowners are now uninsured — many of them seniors who can't afford premiums that have doubled in five years. One storm away from losing everything.

The real jerk signal underneath all of it: insurers have shifted from backward-looking models to forward-looking AI. They used to price risk based on what happened. Now they're using models that update daily, down to the individual parcel and roof type.

When the people whose entire business model is pricing risk decide the past no longer predicts the future — that's jerk.

A fair question: is the 2020s spike a permanent regime change, or a cluster that could revert? The honest answer is that we're still inside the data. But the insurers aren't waiting to find out. They're repricing now. That tells you what the people with the most to lose believe about the shape of the curve.

The meta-signal

In May 2025, NOAA shut down its billion-dollar disaster database — the 45-year dataset that tracked all of this.

Climate Central picked it up. But the signal is clear: the agreement that the government would maintain this data is itself under stress.

The instrument we built to measure the change... stopped measuring.

What my grandfather was doing

He was asking the right question in 1986.

He didn't have enough data yet. The billion-dollar disaster record only went back to 1980. The patterns that are now obvious in 45 years of data were invisible in six.

And he didn't have a way to read what was happening to the acceleration itself.

He had position — the temperatures on graph paper. He was looking for velocity — is it trending? But the data window wasn't wide enough to see acceleration. And he had no way to read what was happening to the acceleration itself.

Now we have the data. And we have the lens.

The count of storms looks flat. That's position. The cost and frequency of destruction is rising. That's velocity. It's rising faster each decade. That's acceleration. The type of destruction, the geography, the seasonality, and the instruments we use to track it are all shifting simultaneously. That's jerk.

Five-minute practice

Go to your insurance carrier's website. Find the loss history or claims data for your county — most carriers publish it, few business owners read it.

Don't look at your premium. Look at the rate of change of claims around you.

Is it linear? Accelerating? Or is the acceleration itself shifting — new categories of claims appearing, new geographies lighting up, new seasons producing damage that used to be confined to summer?

That's the jerk read. And it applies to every risk your business carries — not just weather.