🤖 ChatGDP (Q4: +0.5%)

Another economic "nowcast" brought to you by AI (and it thinks the economy stinks this quarter!)

I recently met with a group of economists and forecasters, and — big surprise — AI was a major theme. How good is AI at economic forecasting? How can it improve existing forecasts? Should AI forecast independently or augment expert humans? All good questions that smart people are studying.

I left the meetup curious about how well an unassisted AI model would generate a “nowcast”. So, with some AI written scripts and a fancy new API key, I set up a code that points Gemini towards a handful of economic news outlets and traditional macroeconomic data feeds. The code has been logging Gemini’s forecasts for a few days now, and so far it thinks that Q4 economic growth will be bad! Over the last 3 days, the GDP growth prediction has averaged +0.5%.

Note that the so-called “Blue Chip” consensus is for +0.8% growth in Q4. I did have to editorialize this output a bit: Gemini seemed to mix up its point forecasts and ranges, so there’s some math behind the scenes to make sense of it. It is interesting that Gemini’s GDP growth prediction fell significantly after Friday’s stock market meltdown. In addition, the forecast range spread grew much wider — possibly Gemini’s way of interpreting the market volatility.

So why is Gemini pessimistic so far this quarter? Here are some of the key words it picked up from the news feeds it is observing:

• Stagnant services PMI

• Deteriorating consumer sentiment

• Weakening labor market

• Slowing retail sales

• Government shutdown

The Prompt

Prompts are very important in the AI world. One of the big challenges is that these models are only trained on data through a specific date (most likely January 2025), so we have to point the model towards news and databases that can help it generate a real-time forecast. In this case, the model is using Google RSS feeds for economic news, the stock market, jobs reports, etc. Here’s what I’m asking Gemini to do:

You are an expert U.S. macroeconomic analyst and forecaster.

Using the following live economic headlines and indicators, generate an independent, data-driven forecast for U.S. real GDP growth (annualized SAAR) for the current quarter. Base your reasoning on qualitative trends from employment, inflation, ISM reports, consumer sentiment, spending data, and the stock market. Do not reference GDPNow or other official nowcasts.

Be explicit about uncertainty:
- Always provide a central forecast (most likely value).
- Include a plausible range that reflects confidence in your estimate.
- The range should normally be at least ±0.3 percentage points wide, and up to ±1.5 if recent data are mixed or volatile.

Output your response exactly in this format:
Forecast: [number]%  
Range: [low% – high%]  
Key Drivers: [short comma-separated list]  
Commentary: [2–4 sentences summarizing reasoning and risks]