For students6 projects6 open

Student research projects

Six research questions in quantitative finance, each sized for one student over six weeks. You do the data, the code and the numbers. I mentor, then write the article and publish it here with you as co-author.

This is free, for anyone who wants to break into quant research.

Every question has a written specification that was locked before anyone claimed it, so the design cannot drift once the results come in. All results get published, including negative ones.

The projects

Who this is for

The projects are pitched at a final-year mathematics undergraduate, or anyone with the same footing: probability, statistics, linear algebra, some time series and enough Python to write tested code. No finance background is needed. Each spec defines the market terms it uses.

If you are not quite at that level but want to try, email me anyway and we can talk about which project fits.

How it works

Pick a project below and click "Claim this project". This opens an email to me.
Tell me a little about yourself: your name, university and year, the project you want and a few lines on your background.
I send you the full specification. It fixes the question, data, methods, the answer rule and the minimum bar for publication. We have a short call to check it is a good fit.
You work on it for six weeks, alone. I review your work at three checkpoints (end of weeks 1, 3 and 6) and answer questions in between. I will help you when you are stuck, but I will not write code or run analyses for you.
You hand in, I write. If the work meets the bar, I write the article and you check it for factual errors. Then it goes up on this site with your name on it.

One student per project at a time.

What you hand in

A code repository that rebuilds every number from the raw data with one command. It stays private, shared with me, while you work, and becomes public under the MIT licence when the article is published.
A results pack: the tables and figures the spec lists, plus a machine-readable summary of the headline.
A methods log of two to four pages: what you did, every deviation from the spec with its reason, plus anything that surprised you.

You do not write the article. That is my job.

What you get

Co-authorship on a published research article, with your university if you wish.
Once the article is out, an open-source repository with your name on it, built to a standard you can show an employer or a supervisor.
Six weeks of one-to-one mentoring on a real research problem.
A reference letter, at my discretion, if the work is good.

The projects are unpaid and voluntary.

The projects

P1Open

Does intraday data beat daily data for forecasting tomorrow's volatility?

For crypto perpetual futures, does a forecast built from five-minute returns predict next-day volatility better than one built from daily returns alone? You compare an HAR model on realised variance against GARCH(1,1) and a simple exponential average on ten major contracts, with formal forecast-comparison tests.

You will produce: out-of-sample forecast losses for three models on ten contracts; Diebold-Mariano tests per contract and pooled, with a multiple-testing correction; a model confidence set; a signature plot; every forecast.
Background: maximum likelihood, OLS asymptotics, stationary time series, HAC variance estimation.
Data: Binance public futures data, one-minute and daily, 2021 to 2026.

Claim this project

P2Open

How fast do perpetual funding rates mean-revert, and is that enough to trade?

When a perpetual futures funding rate jumps away from its usual level, how many days does it take to close half the gap? Does knowing that half-life let a spot-and-perpetual carry position beat random timing after costs?

You will produce: bias-corrected half-lives with confidence intervals for twenty contracts, overall and by year; a simulation check of the bias correction; three carry rules with profit split into funding, basis and costs; a random-entry null test; a full trade ledger.
Background: AR processes, OLS asymptotics and small-sample bias, bootstrap for time series.
Data: Binance public funding-rate history, futures and spot prices, 2021 to 2026.

Claim this project

P3Open

Does the bootstrap you choose decide whether a strategy looks significant?

Trading returns have heavy tails, volatility clustering and sometimes serial correlation. Which of six standard methods for a Sharpe-ratio confidence interval keeps its stated 95% coverage under those conditions, and does the choice of method change the verdict on real trading rules?

You will produce: coverage, interval width, size and power for six methods on five simulated return processes at three sample sizes; a block-length sensitivity check; verdicts from every method on ten classic Bitcoin trading rules, and how many rules flip.
Background: CLT and the delta method, stationary processes, Markov chains, the bootstrap principle.
Data: simulation, plus Binance public Bitcoin spot prices from 2017.

Claim this project

P4Open

How many strategies can you try before the best one is noise?

A researcher tests N trading rules on past data and keeps the one with the best in-sample Sharpe ratio. As N grows, does picking the best one still beat picking at random out of sample? If the benefit disappears, at what N? Do the Deflated Sharpe Ratio and Hansen's SPA test warn about it from the in-sample data alone?

You will produce: in-sample and out-of-sample Sharpe ratios for 1,000 rule configurations over 23 walk-forward windows; the gain from picking the best as N grows, with intervals; pass rates for both tests and whether they predict out-of-sample results; a control run on data with no signal.
Background: order statistics and Gaussian extremes, eigenvalues of correlation matrices, hypothesis testing, the bootstrap.
Data: Binance public Bitcoin futures prices, hourly, 2020 to 2026.

Claim this project

P5Open

Does shrinking the covariance matrix improve a minimum-variance crypto portfolio?

When the number of assets is not small relative to the number of observations, the sample covariance matrix is badly conditioned and its inverse amplifies noise. Do Ledoit-Wolf shrinkage and random-matrix cleaning lower the realised risk of a minimum-variance portfolio, and how does the answer change with the ratio of assets to observations?

You will produce: realised volatility for six estimators across twelve combinations of portfolio size and estimation window; variance tests against the sample covariance; predicted versus realised risk; turnover and net return; a simulation compared with exact finite-sample theory.
Background: spectral decomposition and matrix inverses, multivariate normal theory, Lagrange multipliers.
Data: Binance public futures prices for every contract, including delisted ones, 2021 to 2026.

Claim this project

P6Open

How much does estimation error cost a Kelly bettor?

The Kelly fraction maximises long-run growth when the return distribution is known, but a real trader has to estimate it. If you bet a fraction c of the estimated Kelly fraction, which c maximises realised growth? You derive a closed-form answer, test it by simulation and then measure what happens on real crypto returns.

You will produce: a written derivation of the optimal fraction; a simulation grid with ruin rates; the simulated optimum against the theoretical one; growth curves on ten real contracts after costs and funding.
Background: expectation and variance of sample means, Taylor expansion, logarithmic utility.
Data: simulation, plus Binance public futures prices and funding, 2021 to 2026.

Claim this project

Locked before anyone started

Each specification was locked on 16 October 2026, before any student was assigned. I hashed every file with SHA-256 and timestamped the hashes on the Bitcoin blockchain with OpenTimestamps. When an article is published, the full spec goes out with it, so anyone can check that the study answered the question it set out to answer, in the way it said it would.

Common standard	806cc9ca2bca769c4babf74899b9a4ff25473541d80dfdca53fd7079fc6aa270
P1	9d18cf7bd9b7f81066e4507425126495ccfd0d076a313e486ae4a7558feaca7a
P2	a56a81cdec52cd6ebad02fcf1cf70ee82726c00eeadb389e1405c53a0753b662
P3	a39e6c348e5b08ea11fd62350b191b9d27ae121e58b8bd5d929d7c0731047ba6
P4	10e405c3a9bae3a3a27c381a76e95ba087b7c7c95a1172e75aecc4a4c024d853
P5	f0bdebd01b8698893ffe7a8b66e3a01eae5a16a90e85fc6cdb1cc5c89664c43f
P6	d0583c739f64793f1b168df424789c5992d8e212ff2d055ce6bb10d3a0f7c469

Once locked, a spec does not change. If something in it turns out to be impossible as written, the change is logged with a date and a reason and published with the results.

Terms (summary)

By claiming a project you agree to these terms. The full terms come with the specification.

Unpaid and voluntary. This is not employment or an internship.
Co-authorship needs the bar. If your work meets the minimum bar in the spec, you are a co-author. Partial work is acknowledged by name.
All results are published, positive, negative or null. The direction of a result has no effect on whether the work is accepted.
I write the article and have final editorial control. You see the draft first and can correct any factual error about your work. You can ask to have your name removed, but the results and code may still be published.
Keep results private until the article is live, including your repository. After that, present them, put them on your CV and build on them.
AI tools are allowed, but you are responsible for every line and every number. I may decline work that was clearly produced without understanding, such as code you cannot explain. At any checkpoint I may ask you to walk me through part of the code and re-run it live.
Your code, open licence. The repository is private while you work and goes public under the MIT licence with the article. You keep the copyright.
Public data only. Only the free data named in the spec. No paid, proprietary or employer data.
Integrity. Fabricated or selectively reported results end the project, and any affected article is retracted.
Tell me about conflicts up front, such as working at a trading firm or using the project for university credit.
Claims lapse. If a checkpoint is more than two weeks late without notice, the project can go to someone else.
Not investment advice. The strategies studied are research objects, not recommendations.
What gets published about you: your name, your university if you wish and your repository link. Emails stay private.

Questions before you claim? Email daniel@daru.finance.