Freddie Mac Interview Questions & Experiences

Freddie Mac·Data Scientist·Technical Phone Screen

Jul 2026

Freddie Mac data scientist interview, heavy quant focus on panel econometrics for real estate data. The whole thing felt like a graduate seminar more than a job interview, which I was not fully prepared for.

You have monthly panel data on rental properties across MSAs from 2010 to 2025. Write a regression to estimate the elasticity of rent with respect to vacancy, including property and month fixed effects. Justify your choice of log transforms and any seasonal terms.
Given serial correlation and heteroskedasticity in your panel data, which standard error estimator would you use and why? Compare two-way clustering by property and month, Driscoll-Kraay, and Newey-West. What are the finite-sample tradeoffs?
Test for unit roots and cointegration in log rent and log vacancy. If both series are I(1), how would you set up an error correction model using either the Engle-Granger or Johansen approach to avoid spurious regression?
If your model returns a coefficient of -0.35 on log vacancy, what does that mean? Specifically, compute the predicted percent change in rent when vacancy increases from 8% to 10% at the MSA level.
How would you diagnose multicollinearity among macro regressors like unemployment, CPI, and interest rates? What remedies would you consider, such as orthogonalization, ridge regression, or Bayesian priors, and how does each affect your inference?

“I knew the log-log setup cold, elasticity interpretation is straightforward.”

View Post

Freddie Mac·Data Scientist·Technical Phone Screen

Jul 2026

Technical screen for a data scientist role at Freddie Mac. One question, but it was a beast. Basically a full take-home crammed into a live session covering pandas at scale, feature engineering with leakage constraints, and complexity analysis.

Given a 50-million-row monthly loan panel with columns for loan ID, MSA, month, property type, a 90-day delinquency flag, and unpaid balance, write pandas code to generate three leakage-safe features per loan-month: (a) a 12-month rolling delinquency rate by MSA and property type that excludes the current month, (b) a target-encoded property type delinquency rate per MSA using only data strictly before the current month, and (c) an exponentially weighted default intensity per loan with a 6-month half-life. Then explain how you'd keep memory under 16 GB, guarantee time-order correctness after any shuffles, and unit test edge cases like missing months or single-observation groups. What are the big-O tradeoffs?

“This one took me a second to even parse fully.”

View Post

Freddie Mac·Data Scientist·Online Assessment (OA)

May 2026

Took a Freddie Mac data scientist assessment that was basically a full warehouse audit exercise, schema and all. Eight parts covering everything from deduplication logic to DDL constraints to self-joins. Dense stuff, felt like a take-home that someone decided to give you during a live session.

Given a retail analytics schema with messy data, outline a concrete ordered plan to clean the warehouse: deduplicate customers by email with justified tie-breakers, identify and remove ingestion-duplicate orders using a natural key and time/amount tolerance, standardize data types and time zones, validate referential integrity, and detect impossible values like negative quantities. Provide at least two SQL or Python snippets.
You need to onboard a CSV file with columns for order ID, timestamp, customer email, SKU, quantity, and price in cents into the existing Orders and OrderItems tables. What fields would you include in a formal data mapping specification, and how would you map each CSV column to its target with all required transformations?
Write three queries: one that flags duplicate customers by email and returns only the row to keep with a reason code for dropped rows, one that clusters likely duplicate orders for the same customer where the order date is within 30 seconds and total amount matches to two decimal places, and one that deduplicates products by SKU and picks a survivor deterministically.
Explain the practical differences between DDL and DML in the context of this warehouse schema, including transactional and locking implications. Provide one DDL statement you would actually run here to enforce integrity and one DML statement demonstrating data manipulation.
Using the Customers table, write a self-join query that returns each customer's ID, email, and their referrer's email if one exists, making sure customers with no referrer still appear exactly once.
Show how you would insert or update a product by SKU so that duplicate SKUs collapse into a single correct row without violating constraints. How would you handle concurrent writes?
Define appropriate primary key, foreign key, and unique constraints for all four tables given the sample data. Justify any composite keys and discuss trade-offs between surrogate and natural keys for Customers and Orders.
Rate your SQL proficiency from 1 to 10 and justify it with two concrete examples from the tasks above, such as a window-function-based deduplication and a concurrency-safe upsert. Include how you would test and optimize those queries.

“This was the part that took the longest.”

View Post

Freddie Mac·Machine Learning Engineer·Technical Phone Screen

May 2026

Lead MLE interview at Freddie Mac focused almost entirely on one deep-dive project presentation. You get a few minutes to walk through something you actually owned end-to-end, so if you've only ever been a contributor on a team project, this format will expose that fast.

Walk me through an ML project you led or had significant ownership over, covering the business context, your role, the technical approach, and the results.
What were the key design decisions and tradeoffs you made in building out the pipeline?
What was the biggest challenge you ran into and how did you resolve it?
Looking back, what would you do differently or improve on that project?

“This is the whole interview basically.”

View Post

Freddie Mac·Data Scientist·Technical Phone Screen

May 2026

SQL-heavy interview for a Data Scientist role at Freddie Mac that went pretty deep into query design, performance tuning, deployment safety, and mentorship. Four distinct prompts, each with multiple sub-parts. More of an engineering-flavored screen than I expected for a DS title.

Walk through a complex SQL query you wrote, covering the business problem it solved, how large the data was, and why you chose that specific design over other approaches.
Describe a time you improved SQL query performance. What were the baseline metrics, how did you diagnose the bottleneck, what changes did you make, and what were the results?
Describe how you've safely deployed a SQL migration: code review process, DDL with backfill strategy, feature flags, rollback plan, and how you confirmed data correctness afterward.
How do you mentor others on SQL? What anti-patterns do you watch for in code reviews, and can you give a recent example where your feedback made a query faster or more reliable?

“They wanted real specifics: window functions, CTEs, recursion, something like that.”

View Post

Freddie Mac·Data Scientist·Technical Phone Screen

Apr 2026

Interviewed for a Data Scientist role at Freddie Mac and got hit with a deeply technical ML engineering question about time-series cross-validation for mortgage data. One question, very long, very specific. The kind of thing where you either know the domain or you're fumbling through it live.

Design a panel-aware blocked time-series cross-validation splitter with an embargo period. The input is a DataFrame with loan_id, MSA, and month columns. Requirements include: K=5 expanding-window folds with embargo logic preventing any loan_id from appearing in both train and test within 90 days; grouped blocking so at least one fold holds out an entire MSA as test; determinism under shuffled input; pseudocode plus time/memory complexity analysis; and unit tests covering duplicate timestamps, missing months, loan migration across MSAs, and highly imbalanced MSAs.

“This question is basically a take-home disguised as a verbal question.”

View Post

Freddie Mac·Machine Learning Engineer·Onsite - System Design / Architecture

Apr 2026

System design round at Freddie Mac for an ML Engineer role, focused entirely on a single massive RAG question covering basically every dimension of a production AI system. No coding, no behavioral stuff, just this one sprawling prompt and however long you could hold the thread.

You run an enterprise RAG system for high-stakes document QA like mortgage underwriting or legal review. Users don't trust the answers because they can't verify sources, gauge uncertainty, or know when the model might hallucinate. How do you redesign the system to increase user confidence? Cover the full stack: product UX, retrieval and generation architecture, confidence scoring, calibration, fallback behavior, human feedback loops, monitoring, privacy, compliance, and adversarial testing.

“This question is essentially a whole system design interview compressed into one prompt.”

View Post

Freddie Mac·Data Scientist·Technical Phone Screen

Apr 2026

Freddie Mac data scientist technical screen, heavy SQL focus across about six problem areas. The questions were legitimately hard and covered everything from messy staging table cleanup to upsert concurrency. Felt more like a take-home than a live screen given the depth expected.

Write a single SQL statement to clean a messy staging table: normalize emails to lowercase, strip whitespace, collapse double spaces in names, validate emails by a simple rule, map country strings to ISO alpha-2 codes, parse flexible date formats to TIMESTAMP, and deduplicate by normalized email keeping the most recent row (tie-break on smallest ID).
Provide a source-to-target mapping spec to load the cleaned staging data into the customers table, including data types, transformation rules, nullability, and how updated_at gets populated. Show a concrete before/after example for at least one row.
Find duplicate person records in the customers table using case-insensitive trimmed email as the uniqueness key. Then identify which cleaned staging records would collide with existing customers, showing both sides' identifiers.
For three specific operations, state whether each is DDL or DML, then write the exact SQL: adding a unique constraint on lower(trim(email)) in customers, adding a foreign key from orders to customers with a deletion rule you justify, and creating indexes to support the duplicate-finding and upsert queries.
Using a self-join on the orders table, list pairs of orders by the same customer where one order date is exactly one day after the other. Output customer ID, both order IDs, and both dates. Avoid returning duplicate pairs and think about performance on large tables.
Implement an upsert from the cleaned staging data into customers as a single SQL statement. On conflict, prefer the longest non-null normalized name, update country_code only if it's currently NULL in customers, and set updated_at to now. Explain how your approach handles race conditions under concurrent load.

“This one took me a while to untangle mentally.”

View Post

Freddie Mac·Data Scientist·Technical Phone Screen

Apr 2024

Interviewed for a Data Scientist role at Freddie Mac and got hit with a deep causal inference question about a real FHFA policy change. One question, but it had five sub-parts and basically covered everything from DiD estimators to how you'd explain prediction intervals to an exec. Dense stuff.

The FHFA tightened underwriting standards for multifamily loans on April 15, 2024. Design a quasi-experiment to estimate the causal impact on monthly originations and 12-month delinquency rates. Walk through your treatment and control definitions, unit of analysis, DiD approach versus synthetic control, pre-trend testing, falsification checks, spillover concerns, outcome definitions, standard error clustering, how you'd communicate uncertainty to executives, data needs, and how you'd handle policy anticipation and partial compliance.

“This was a lot.”

View Post