Does a carry signal in Treasury futures survive realistic roll costs?
Carry looks attractive on paper. The question is how much is left after the cost of rolling contracts each quarter, and whether that remainder is stable across rate regimes.
Example questions
Examples of the kinds of questions our interns and new researchers work on. Your own question is agreed with your mentor when you start.
About this page
Every Lenox intern owns one question for the internship. A good one is narrow enough to answer in the weeks available, important enough that a team will act on the answer, and honest about what would count as failure.
The questions below show the range of work across our teams. They are written in the form we use internally: the question, why it matters, and the methods you would likely use. None of them is a description of a current Lenox project or position.
When you apply, you do not need to propose a question. If one of these interests you, say so in your application; it helps us match you with a team.
Quantitative trading
Carry looks attractive on paper. The question is how much is left after the cost of rolling contracts each quarter, and whether that remainder is stable across rate regimes.
If most of a signal’s value arrives in the hours after a revision, it needs different execution from one that unfolds over weeks.
Execution algorithms spread orders according to expected volume. A forecast error becomes a cost; the question is how much a better forecast would save.
Revised data can make a strategy look better than it could have been. Measuring the gap on a known signal shows how much caution our reviews need.
Fixed income
Ratings change slowly. A company paying down debt may be a better credit than its rating suggests before the agencies agree.
Large auctions may push yields up ahead of the sale and down afterward. If the pattern is reliable, it informs when we buy.
Mortgage pools contain borrowers with different habits. Testing a model on repeat refinancers shows where it is weakest.
Asset management and client solutions
Harvesting losses helps after-tax returns but moves a portfolio away from its benchmark. Measuring the trade-off helps clients choose how far to go.
A hypothetical plan with a funding gap needs to hedge its liabilities without giving up too much return. The question is where that balance sits.
Risk and technology
| Question | Team | Likely methods |
|---|---|---|
| How often did our value-at-risk estimate understate losses, and in which markets? | Risk | Model backtesting, statistical tests |
| How many days would it take to sell each holding without moving its price? | Risk | Liquidity data, market impact estimates |
| Can a language model classify company filings more reliably than keyword rules? | Data Science | Natural language processing, time-aware validation |
| Which vendor data corrections would change a research result, and how do we catch them early? | Data Engineering | Data quality checks, pipeline design |
| How much faster can a common research query run on a columnar store? | Data Engineering | Benchmarking, query optimization |
You do not need a question to apply. You need the habit of asking good ones. Tell us which kinds of questions interest you and we will match you with a team.