Nowcasting Inflation Using Online Prices
This study demonstrates that a daily food price index constructed from 150 million online retail entries using transformer-based language models can effectively nowcast official Turkish inflation rates, particularly during periods of low and stable inflation, though the correlation weakens when inflation spikes.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine you are trying to guess the final score of a football match before the referee blows the final whistle. Usually, you have to wait until the game is completely over and the official stats are released to know the true score. In the world of economics, this "final score" is the inflation rate (how much prices have gone up), and the "official stats" come from the government's statistical office (TurkStat), which releases the data a month late.
This paper is about a team of researchers who built a real-time scoreboard using online prices to predict that official score early.
Here is how they did it, broken down into simple steps:
1. The Problem: The "Slow Mail" vs. The "Live Stream"
Official inflation numbers are like a letter sent by snail mail. The government sends teams of people to stores to write down prices, or they scan receipts from physical stores. This takes time. By the time the government publishes the number, it's already a month old.
The researchers wanted a "live stream." They decided to watch the prices on five major Turkish online grocery websites every single day. They collected between 50,000 and 70,000 price tags every day for over seven years. That's a massive amount of data—about 150 million data points in total.
2. The Challenge: The "Messy Library"
Imagine walking into a library where the books are thrown everywhere. Some books are labeled "Food," but inside that pile, you find a book about "Coffee Filters" mixed in with "Espresso Beans." Some items are labeled "Milk," but they are actually "Oat Drinks."
The websites the researchers scraped had similar problems. The categories were messy, inconsistent, or just plain wrong. If they just searched for the word "Coffee," they might accidentally count the price of a coffee filter as the price of coffee beans.
3. The Solution: The "Smart Librarian" (AI)
To fix the messy library, the researchers didn't hire humans to sort every single item (that would take forever). Instead, they built a Smart Librarian using Artificial Intelligence (specifically, a type of AI called NLP or Natural Language Processing).
- Training the Librarian: They first manually sorted about 150,000 items by hand to teach the AI what "Coffee Beans" actually look like versus "Coffee Filters."
- The Result: The AI learned to read the product names and sort them correctly with 95% accuracy.
- The Job: Once trained, this AI librarian went to work every day, sorting 50,000 to 70,000 new items instantly, organizing them into the exact same categories the government uses.
4. Building the "Daily Scoreboard"
Once the AI sorted the prices, the researchers built their own Daily Food Price Index.
- They used the same math and "weights" (importance of different foods) that the government uses.
- Instead of waiting a month, they updated this index every single day.
5. What They Found: The "Weather Forecast" Effect
When they compared their daily online index to the government's official monthly index, they found two interesting things:
- When the weather is calm (Low Inflation): The two scores matched up perfectly. If prices were stable, the online prices looked just like the offline prices.
- When the storm hits (High Inflation): When inflation started to skyrocket (reaching 85% in Turkey), the two scores started to diverge. The online prices moved differently than the official offline prices.
- The Metaphor: It's like how a weather app might predict rain based on satellite data, while people on the ground are still looking at a clear sky. During chaotic times, the "live stream" (online) and the "official report" (offline) tell different stories.
6. The Big Win: Predicting the Future (Nowcasting)
The most important part of the paper is Nowcasting. This means predicting the official number before the government releases it.
The researchers used their daily online data, combined with other economic clues (like exchange rates and interest rates), to guess what the official inflation number would be at the end of the month. They tested this using many different computer models (like a team of different forecasters).
- Early in the month (Day 2 or 15): The online data was a bit "noisy" and didn't help much. It was like trying to guess the final score of a football game after only 10 minutes of play; the data was too chaotic.
- Late in the month (Day 22): Once they had collected data for most of the month, the online prices became a powerful crystal ball. Adding the online data to their models made their predictions significantly more accurate.
The Bottom Line
The paper concludes that while online prices are great for getting a quick peek at inflation, they are most valuable when used to predict the official number late in the month.
However, there is a catch: When inflation gets very high and chaotic, online prices and physical store prices start to drift apart. This suggests that during crazy economic times, the government might need to rely more on online data to get a true picture of what is happening, rather than just sticking to their old methods.
In short: They built a super-fast, AI-powered price tracker that can guess the government's inflation report a month early, but it works best when the economy is a bit settled and the month is almost over.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.