Mortgage Market Forecasting With the Model Match MCP

Mortgage market forecasting in the Model Match MCP projects purchase and refinance loan activity up to 12 months ahead, with a low-to-high range and a mortgage-rate scenario you choose, for a state, county, metro, or ZIP code. You ask in plain English in Claude or ChatGPT, and the answer comes from the same mortgage market intelligence behind Model Match Market Signals.

Most mortgage teams plan from the rearview mirror: last quarter’s units, last year’s volume. What a goal, a staffing plan, or a board slide actually needs is a range for the months ahead and an honest answer to “what if rates move?” A single number can’t give you either. A forecast that comes with a band and a rate assumption you control can.

What a mortgage market forecast includes

Each forecast returns three things, and you can slice it by state, county, metro, or ZIP code and by loan purpose, loan type, property type, first-time homebuyers, veterans, and more:

• A likely number. Monthly loan units, dollar volume, or average purchase price, one to twelve months out.
• A low-to-high range. The 10th and 90th percentiles around the likely number, so you can plan for the downside and the upside.
• A rate scenario. Rates are an input, not a prediction: hold them flat, move them down or up one point, or supply your own path.

The prompt

Copy this into Claude or ChatGPT with the Model Match connector turned on. Replace the items in brackets with your own market and measure:

What comes back

We ran that prompt on September 26, 2026. Here’s the forecast for the months still ahead, from October onward — the solid line is the likely number, the shaded band is the 10th-to-90th-percentile range:

Month	Rates flat: likely (low–high)	Rates fall 1 point: likely (low–high)
Oct 2026	454 (368–541)	472 (388–552)
Nov 2026	400 (311–485)	420 (336–502)
Dec 2026	370 (271–463)	395 (301–480)
Jan 2027	470 (371–566)	495 (402–583)
Feb 2027	532 (430–627)	560 (466–647)
Mar 2027	633 (537–720)	661 (575–741)

Purchase originated units, Mecklenburg County, NC. Figures are rounded and change as new data arrives.

Three things stand out, and they’re the reason to ask for a forecast instead of a trend line:

• It keeps the seasonal shape. Purchase originations dip to a December low, then climb into a spring peak, the same rhythm the county shows in its history.
• The rate scenario has a size. If rates fall a point over the year, the model puts these six months about 5% higher, roughly 143 more originated loans (3,003 versus 2,859).
• The range is the planning tool. December’s likely 370 loans comes with a range of 271 to 463. Staff for the low; make sure you can absorb the high.

Mortgage market forecasting for loan officers

A loan officer’s real question is where should I spend my time? Pair a forecast with a loan officer production lookup and you can see where your own numbers stand next to where the market is heading. Here is what loan officers ask:

• Pick your farm area. “Forecast purchase loan volume for the next 6 months in Mecklenburg, Cabarrus, and Union counties. Which one is set to grow fastest?”
• Prep for realtor meetings. “What’s the 6-month outlook for purchase activity in zip 28601? Give me 3 talking points I can use with agents this week.”
• Build a rate-drop playbook. “If rates fall a full point over the next 6 months, how much does refinance volume in North Carolina go up versus rates holding flat?”
• Focus your products. “Forecast FHA and VA units in the Charlotte metro for the next 6 months. Which is growing faster?”
• Time first-time buyer marketing. “Project first-time homebuyer purchase activity in my three core counties through spring. When should I ramp up my FTHB marketing?”

Forecasting for branch and sales managers

Managers plan and coach against a number. A low, likely, and high case gives you a goal you can defend and a way to spot capacity problems before they show up in turn times.

• Set realistic goals. “Forecast originated units in our branch footprint (these 5 counties) for Q4 and Q1. Give me a low, likely, and high case for goal setting.”
• Plan staffing and capacity. “Based on the 6-month purchase forecast for Dallas-Fort Worth, will we need more processing capacity heading into spring?”
• Build team meeting content. “Build me a one-slide market outlook for my Monday meeting: 6-month purchase and refi forecast for Arizona under flat rates and falling rates.”
• Make territory decisions. “Compare the 6-month forecast for purchase volume across the top 10 metros in Florida. Where should I add my next loan officer?”

Scenario planning for mortgage executives

Expansion, budgeting, and rate exposure are scenario decisions, and a range with a rate assumption you control maps to how boards and planning committees already think. Pair a forecast with a competitive view by building a mortgage expansion strategy with Claude.

• Run scenarios. “Run the 12-month refinance forecast nationally under three rate scenarios: hold, fall, and rise. Summarize the spread for a board slide.”
• Decide where to expand. “Which 5 states are forecast to grow purchase volume the most over the next 12 months? Show the range, not just the midpoint.”
• Use your economist’s outlook. “Our economist expects rates at 6.2% in December, 5.9% by March, then flat. Forecast refi units in Texas over the next 12 months using that path.”
• Watch channel and mix shifts. “Forecast units by loan purpose (purchase vs refi) for the next 6 months in our top 3 states. Is the mix shifting?”
• Set community lending goals. “Forecast purchase originations in low- and moderate-income neighborhoods in our assessment area for the next 12 months.”

Mortgage data analysts: scenarios, ranges, and tables

Analytics teams get the most from the parts of a forecast that are easiest to skip: the range, the rate path, and side-by-side comparison. Each response includes the model used (Amazon Chronos-2, a zero-shot time-series model), the as-of month, the rate path, a confidence label, and the 10th, 50th, and 90th percentile for every month, so results can be checked and dropped into your own models. Ask for tables.

• Build a scenario grid. “Forecast purchase originated units for the next 12 months in Texas under flat, falling, and rising rates. Return one table with the likely, low, and high value for each month and scenario.”
• Test how geography affects stability. “Forecast purchase originated units for the next 6 months in zip 28601 and in the county it sits in. How wide is the low-to-high range in each?”
• Measure the application-to-close gap. “Forecast purchase units and purchase originated units for Arizona over the next 6 months. How large is the gap between applications and closed loans?”
• Compare loan type mix. “Forecast FHA, VA, and conventional purchase originated units in the Charlotte metro for the next 6 months. Return a table with the likely, low, and high case for each.”

Metro filters only include loans that carry a metro code, so a metro forecast can be narrower than the state or county around it.

How rate scenarios work

A forecast is conditioned on a 30-year mortgage rate path, which turns “what will happen?” into “what happens if?”

Scenario	What the forecast is given	Ask it like this
Hold	The latest weekly Freddie Mac 30-year rate, held flat (default)	“If rates hold steady…”
Fall	The rate gliding down one point, evenly, over the horizon	”If rates fall a full point…”
Rise	The rate gliding up one point, evenly, over the horizon	”If rates rise a full point…”
Custom	Your own month-by-month rate path	”Rates at 6.2% in December, 5.9%…”

How to read a mortgage market forecast

• Read the range, not just the middle. The low is the 10th percentile and the high is the 90th. Goals and staffing belong closer to the low; the high is what you need capacity for.
• Expect wider ranges in smaller areas. For June 2026 purchase originations, the range was about a quarter of the likely number in Mecklenburg County, about 44% in Catawba County, and about 89% in ZIP 28601, which the tool also labeled medium confidence instead of high. A county or metro gives steadier results than a single ZIP code.

• Know where the forecast starts. It begins after the latest month with complete data. Recent originations are still being reported, so our September 2026 run for originated loans began in June 2026. That is why the table above starts in October, and why asking for 12 months gives you a longer look ahead.

A forecast is an estimate, not a guarantee. Use it next to your own pipeline and local knowledge, and rerun it monthly as new data arrives.

How to connect Model Match to Claude or ChatGPT

Forecasting runs through the Model Match MCP connector. Setup happens once and takes about a minute:

1. Open your Model Match Connectors settings, then add the Model Match connector in Claude under Settings → Connectors.
2. Sign in with your Model Match account and approve access.
3. Start a chat, confirm the connector is on, and paste a prompt from this guide.

There are no API keys to manage and nothing to reinstall. For step-by-step screenshots, see the help-desk guide on connecting Model Match to Claude, or connecting Model Match to ChatGPT.

Forecasting is one part of what the connection does. The same setup answers who’s producing in a market, builds a loan officer market position report, and explains what the Model Match MCP server is if you’re new to it.

FAQs

What is mortgage market forecasting in Model Match? Mortgage market forecasting in Model Match projects purchase and refinance loan activity, such as originated units, dollar volume, and average purchase price, up to 12 months ahead. Each forecast returns a likely number, a low-to-high range, and follows a mortgage-rate scenario you choose. You run it by asking in plain English through the Model Match MCP in Claude, ChatGPT, or another AI assistant.

How far ahead can Model Match forecast mortgage volume? Up to 12 months, with one value per month. The horizon counts from the latest month with complete data, not from today. Recent originations are still being reported, so in a September 2026 run for originated loans the forecast started in June 2026. Ask for 12 months when you want a longer look ahead.

What do the low and high numbers in a forecast mean? For each month, the low is the 10th percentile of the forecast, the high is the 90th percentile, and the likely number sits in the middle. Plan goals and capacity around the low, and make sure you can handle the high. Smaller areas produce wider ranges than larger ones.

How do mortgage rate scenarios work in a forecast? The rate is an input the forecast conditions on, not something it predicts. The default holds the latest weekly Freddie Mac 30-year mortgage rate flat. The fall and rise scenarios move that rate down or up one point, evenly, across the forecast horizon. You can also supply your own month-by-month rate path, such as an economist's outlook.

What is the difference between units and originated units? The default units and volume measures count loan applications of every outcome, including denials and withdrawals. Originated units and originated volume count closed loans only. Ask for originated units whenever you want production numbers for goal setting, staffing, or comparison with funded loans.

How accurate are the forecasts? A forecast is an estimate, not a guarantee, which is why it comes with a range and a confidence label. Broader areas such as counties and metros give steadier results than a single ZIP code, and forecasts work best with 12 to 24 months of history. In our runs, a county forecast was labeled high confidence while a single ZIP code was labeled medium, with a range about twice as wide. Use forecasts alongside your own pipeline knowledge.