Jonathan Rueffer
Science Editor
On Thursday, March 5, Assistant Professor of Statistical and Data Sciences Changzhi Ma presented her research in Taylor Hall as part of the Department of Mathematical & Computational Sciences (MCS) Colloquium series. Her talk, titled “How does inflation remember the past?” examined how inflation is modeled using data from the Consumer Price Index (CPI).
Ma said she enjoyed connecting the research topic to something many students experience in real life. “I had a lot of fun giving this talk because I could start with something many students have noticed lately, prices going up, and then use the data to make sense of it,” said Ma.
Ma began by introducing CPI as a widely used economic measure that tracks the cost of a standard basket of goods and services. Changes in the CPI over time create the inflation rate, which reflects how quickly prices are rising. To clarify the relationship, Ma compared the two concepts by describing CPI as the position on a road, while inflation represents the speed. Using this analogy, she explained that prices can be high overall while the inflation rate remains relatively low.
“I think Dr. Ma did a good job explaining the more complex parts of her statistical model with good metaphors and examples for those who did not have prior knowledge,” said Charles Cain ’26.
CPI data is publicly available through the U.S. Bureau of Labor Statistics. Using their graphs, Ma highlighted the increases in prices during the COVID-19 pandemic. Since CPI data is recorded sequentially over time, it belongs to a category known as time-series data, which tracks measurements at regular time intervals. Ma then explained that CPI time series data involves two major challenges for researchers: non-stationarity and long memory.
A time series is called stationary if it behaves similarly over time, such as its average and variability. However, prices generally rise over long periods, meaning that CPI data is non-stationary and the average value changes over time. As a result, CPI behaves differently today than it did decades ago.
To address this issue, Ma described a method called differencing, which focuses on changes in CPI rather than the overall price level. For example, by looking at month-to-month changes in CPI, the data becomes more stable and subsequently easier to model.
The second challenge involves how strongly past values influence present ones. Some processes “forget” past changes quickly, whereas others retain their effects for longer periods. This latter property of a time series is known as long memory.
To study this effect in CPI data, Ma used a method called the autocorrelation function (ACF), which measures how closely a dataset resembles itself after a delay in time. If the correlation is large, it implies that the past can help predict the present.
Ma’s research investigates how different estimations of the long-memory property can improve predictions of inflation. Her findings suggest that accounting for long memory significantly reduces forecasting error when predicting CPI.
Huy Phan ’26 said that the mathematical modeling offered a “more intuitive and interesting way to think about science, in addition to the incredible applications of it.”
In the final part of her presentation, Ma compared this theory-driven approach with a data-driven machine learning approach. She used a random forest model, which is able to capture complex nonlinear patterns in data but is often less transparent than statistical theory-based methods. Ultimately, she concluded that the two approaches can complement each other for effective predictive modeling.
Ma said the MCS Colloquium series cultivates an environment “where students can take a real problem people care about and talk through the ideas behind it in a way that invites questions and discussion. It gives students a chance to try out their thinking, ask questions, and see different approaches in a supportive setting.”