---
title: "Measuring the economy with new instruments: scraped prices, card-spending trackers and satellite imagery"
kind: "Article"
topic: "Prices & markets"
date: "2026-10-10"
url: "https://canopticon.com/articles/measuring-the-economy-with-new-instruments"
author: "Chris Telford"
license: "CC BY 4.0"
---
Every few years someone announces that enough data and enough computing power will finally let us run the economy properly. It has not happened, and the reasons it cannot are old and well understood. Meanwhile the same technology has quietly transformed something less glamorous and far more useful.

## Why prediction keeps slipping out of reach

Four separate arguments say an economy resists being computed, and they are not about insufficient data. The most familiar is the one about indicators: <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">When a measure becomes a target, it ceases to be a good measure</q>. Its social-science version is broader — <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">the more any quantitative social indicator is used for social decision-making, the more subject it will be to corruption pressures</q>.

Then there is the problem that beliefs about the system alter the system. As George Soros puts it, <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">if investors believe that markets are efficient then that belief will change the way they invest, which in turn will change the nature of the markets in which they are participating</q> — a loop he concedes is <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">difficult, if not impossible, to model</q>. Models do not merely describe markets either; sometimes they build them. The options-pricing model did not document an existing trade: <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">in 1970, there was almost no trading in financial derivatives such as 'futures.' By June of 2004, derivatives contracts totaling $273 trillion were outstanding worldwide</q>.

The best-known version in economics is the Lucas Critique: <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">The parameters of an econometric model estimated under one policy regime won't hold under a new regime, because rational agents form expectations about policy and change their decision rules.</q> A whole school, complexity economics, treats the economy as <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">a non-equilibrium complex adaptive system of heterogeneous, imperfectly-informed agents</q>, and uses simulation to study instability rather than to compute answers. The sociologist Donald MacKenzie summed up the performative point in a phrase: <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">Economics is "an engine, not a camera."</q> The same trap appears inside AI itself; the research notes that <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">reward-model over-optimization in LLM alignment is explicitly described in the research literature as a Goodhart's Law problem</q>.

## What computation actually won

Set prediction aside and the record since the financial crisis is striking. A pair of economists started scraping retail websites and built an inflation index out of them: <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">By 2010, we were collecting 5 million prices every day from over 300 retailers in 50 countries</q> — against the ~80,000 prices the American statistical agency was collecting each month. The product of that scraping is <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">published daily, rather than monthly, and with a lag of three days, rather than several weeks</q>, and it exposed a government that was falsifying its own inflation figures. Its builders were explicit about the choice behind it: they <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Key findings">focus on measurement instead of prediction</q>.

The same pattern holds for the pandemic's best economic instrument, a public dashboard assembled from card-payment and payroll records, which reported spending and employment at county level <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">with a lag of "three days or less."</q> Neither project forecast anything. They shortened the distance between an event and the public's knowledge of it.

The scraped index also caught a government in the act. It <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">famously exposed Argentina's manipulated official CPI</q>. The lesson, in the research's words, is that <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">the value was measurement honesty and speed, and it was buildable by scraping public web prices</q>. Canada's own delay is partly a choice; Statistics Canada <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">has acknowledged it could release faster but core users (Finance, BoC) preferred fewer revisions over speed</q>.

## What AI adds: turning text and images into numbers

The genuinely new capability is narrower than the hype and more useful. <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Key findings">AI's genuinely new capability is turning unstructured text and imagery into structured economic signal</q>. Central banks are already doing it. At the ECB, researchers showed <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">ChatGPT-generated sentiment from PMI news releases improves GDP nowcasts</q>, and at the St. Louis Fed, <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">LLMs can outperform survey forecasts of inflation</q>. One economist used the method on old newspapers, reading expectations out of the Wall Street Journal and <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">extending an expectations series back 120 years</q>.

Images work the same way. Night-time satellite lights <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">correlate with GDP and have been used to estimate subnational and informal-economy activity where official data is absent</q>. The hype is in the opposite direction, toward simulating whole economies. Simulators that test tax policy on artificial agents are real research tools, but they <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">explicitly make "strong assumptions" (static skills, instantaneous labor response, small populations)</q>, and treating their outputs as prescriptions <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">walks directly into the Lucas Critique and performativity problems</q>

## What is still not counted

Large parts of ordinary economic life are missing from the official picture. Informal work is the clearest case: <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">as of 2015 roughly 37 percent of nonretired U.S. adults participated in some type of informal work</q>, and counting it properly would have made <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">the employment-to-population ratio... 0.25 to 1.1 percentage points higher over the 2015–2022 period and as much as 5.1 percentage points higher under more generous estimates</q>. For gig work the estimates do not even agree on an order of magnitude — <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">Global online gig-worker counts range from 154M to 435M depending on definition — the range itself is the measurement failure.</q>

Some of it is not hidden but structurally uncountable by the instrument in use. National accounts measure what is paid for, so a free replacement for a paid product registers as a loss: <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">replacing paid Britannica with free Wikipedia lowers GDP while raising welfare</q>.

The free-goods gap can be put in numbers, by one adjusted measure: <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">Including Facebook's welfare gains would have added 0.05-0.11 points/year to US GDP-B growth.</q> And one quantity that matters to every town is measured almost nowhere: <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">There is no regularly-published official statistic — Canadian or otherwise — for how much money recirculates within a local economy.</q> The studies that try produce striking numbers that do not travel well: <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">These figures vary widely by place and method, rest on small samples, and Civic Economics itself warns against over-extrapolation</q>

## Who owns what: a registry with holes

Hidden ownership is the clearest case where the data exists and nobody can read it. The leaked-documents database <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">exposes "more than 810,000 offshore entities... in more than 200 countries and territories"</q>, but linking leaks to registries and real assets <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">is still largely manual</q>.

Canada has its own reputation here, as a <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">"corporate transparency backwater" ("snow-washing")</q>. A federal register opened for filing in 2024, but it covers a minority of firms: <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">only ~14.5% of Canadian private companies are federally incorporated — the rest are provincial, and only BC and Québec have registries</q>. The research's verdict is that <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">the national picture is fragmented and full of holes.</q>

## How much the top holds depends on who counts

Canada's official wealth figures stop at a level that hides the top. Statistics Canada <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">publishes wealth only at the quintile level</q>, and in late 2025 the top fifth <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">was 65.7%</q> of net worth. Its benchmark survey says plainly that <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">"the share of wealth held by the top 1% will be understated in this data source."</q>

When the budget office built its own database, the number more than doubled: <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">"the top one per cent of economic families own around 23.8 per cent of net wealth in Canada versus the 11.0 per cent suggested by the 2023 SFS"</q>. Another estimate traces the top share <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">rising from 19.3% (1999) to 22.7% (2023)</q>. The research treats the disagreement as the finding: <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">The gap between official and corrected figures is itself a story and a dataset.</q>

## Procurement and lobbying: published, but unreliable

Federal contracts and lobbying contacts are published as open data. That is not the same as legible. One analyst calls the federal open-data portal <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">"a repository of often unreliable data"</q>, notes that <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">the statutory five-year Lobbying Act review has been ignored since 2012</q>, and points out that <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">Canada (unlike the US) doesn't require disclosure of lobbyist spending</q>.

## The instruments exist; the public does not have them

Where new measurement has been built, it has mostly been built privately. Counting cars in retail car parks from satellite images is a real technique with a measured payoff — <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">the strategy yielded a 4-5% return in the three days around quarterly earnings announcements</q> — and the research noted that <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">the edge stayed with a few large hedge funds, not the public</q>. That is an industry now: <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="TL;DR">USD 11.70 billion in 2025</q> of alternative data sold to institutions.

The public version lags accordingly. In Canada the headline series arrive <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">with a delay of six to eight weeks</q>, revised repeatedly, and the central bank <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">doesn't publish its nowcast</q>. The gap is not technical. The same methods that price a hedge fund's quarter could date a recession for everyone else, and nobody is being paid to do that.

Even the size of the private market is uncertain; another firm <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">puts it at USD 18.74 billion</q>, a gap the research attributes to scope. Adoption is not in doubt: <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">"67% of professionals already deploy alternative data and 94% plan higher outlays in 2025,"</q> in one industry survey.

The one place where the books are open is the one marketed as private. On public blockchains every transaction is visible, which produces an irony: <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">crypto, marketed as private, is in many ways the most measurable financial system ever built, while the legacy banking system remains the black box.</q> Even there, <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">it "doesn't capture off-chain activity" (centralized-exchange trades)</q>.

> **Objection: Models are getting better; prediction will come**
>
> The barriers above are not accuracy problems that a larger model dissolves. A policy rule estimated under one regime stops holding when the regime changes and people adjust to it; an indicator decays once it is used for control; a belief about the market changes the market. Simulators of whole economies are <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="TL;DR">real research tools but not oracles</q>. None of that touches measurement, which is a question about what already happened.
>
> Measurement has its own version of the trap, and the research names it: <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Caveats">Any successful public instrument that becomes a target will eventually be gamed (Goodhart).</q> The defence is method, published: <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Caveats">Design for honesty and transparency of method, and expect the measure's usefulness to erode if it ever drives behavior.</q>

## What this explains

It explains why the most consequential economic work of the last decade looks unglamorous — scraped prices, card records, satellite pictures of car parks — and why none of it is a forecast. The gain in each case was time and honesty: knowing sooner, and knowing when an official number was wrong. It also explains where the next one is likely to come from, which is not a better model of the economy but a better view of it. As the research puts it: <q doc="the-telescope-problem-where-ai-and-real-time-data-make-the-invisible-e" section="Details">The telescope didn't predict where planets should be; it showed where they were.</q>