Legislative fiscal staff unveil new Medicaid forecasting model to check agency estimates

Josh Pollet of the Legislative Fiscal Division introduced a presentation to the House Human Services Committee on the division’s rebuilt biennial Medicaid forecasting model. The presentation, delivered by a Fiscal Division analyst identified in the hearing as Ms. Hamilton, framed the model as a legislative check on the executive branch’s Medicaid budget request and said the program accounts for about $5,100,000,000 of the state budget.

Ms. Hamilton said the new model was reconstructed from the ground up to incorporate Medicaid expansion, reduce manual adjustments and scale to roughly 200 different provider time series. “We use this to forecast those short term biannual Medicaid expenditures for the legislature,” she said, describing the model as data‑driven, method‑driven and quality‑driven.

Hamilton outlined the primary data sources: provider claims and reimbursement files (identified in testimony as “901 reports”) available from the data warehouse, extracts from the statewide accounting system Sabers for provider types not in the 901 reports, and macroeconomic projections from S&P Global. To handle incomplete recent claims data, the team applies a percent‑completion adjustment based on the historical lag between service and final reported reimbursements.

On methods, Hamilton said the office runs multiple candidate approaches, using an ARIMA model as a baseline and selecting the best performer for each provider type by fit metrics such as root mean squared error and R². “If none of the models are better than that ARIMA, it goes back to that ARIMA,” she explained, but noted that in many cases exponential smoothing or other approaches produced smaller errors for particular provider types.

The division demonstrated interactive Power BI rollups that combine historical data (back to 2004 for many providers) and forecasts through 2028, with the ability to view projections by program area and provider type. The Fiscal office presented consolidated projections for the House Bill 2 budget using an assumed FMAP of 61.7% in fiscal 2026 and 61.47% in 2027 and noted state special revenue funds were being held near $130,000,000 in the rollup; most projected increases would affect the general fund.

Hamilton also compared LFD projections to Department of Public Health and Human Services estimates, reporting differences in the tens of millions of dollars (LFD lower by about $61.1 million in 2026 and about $78.1 million in 2027 on traditional Medicaid, and by roughly $31.2 million and $45.1 million on expansion in the two years shown). She cautioned that the executive branch’s February projections typically converge with legislative estimates as newer data become available.

Representative Zephyr asked whether machine‑learning approaches (deep AR, H2O) would be adopted. Hamilton said experiments are ongoing but computational cost and parameter optimization currently limit their operational use.

The committee did not take action on the presentation. The Fiscal Division said the model and interactive visuals appear on the Section B subcommittee webpage for review.