Students considering a summer at a small trading firm often ask the same question: what will I actually do for twelve weeks. The honest answer is that you will own one research question from start to finish, and your work will be judged on whether the answer is defensible rather than whether it is impressive. That is the entire shape of the Optilic internship, and it is close to the shape of similar programs elsewhere.
What follows is a realistic example project. The details are illustrative, not a description of any specific live strategy, and the point is the structure rather than the numbers. If you are evaluating whether an internship like ours is a fit, this is roughly what the twelve weeks look like from the inside.
Week one and two: the question
A supervisor hands you a loose version of a question. Something like: does opening auction imbalance in ASX equities predict the direction of the first thirty minutes of continuous trading, and if so, is the signal strong enough to be worth acting on after costs?
Your first job is to make this precise. What exactly is "imbalance"? Signed share quantity, dollar quantity, or a ratio? Measured when, relative to the auction? Predicting what, exactly: return, direction, realised volatility? Over which window? Half the value of the project is in these decisions, and a sloppy specification quietly ruins everything downstream. You write a one-page brief and your supervisor pokes holes in it. You revise.
Week three and four: data
You pull historical order book snapshots and auction messages for a universe of a few hundred ASX names. You immediately discover that the data is not what you expected: a corporate action window has strange prints, three symbols changed ticker during the sample, and the exchange's auction imbalance field has a definition change part-way through. You spend most of two weeks on this. New interns are often surprised by the ratio; experienced researchers are not. The data cleaning is the project, more than the model is.
You build a reproducible pipeline: raw data in, cleaned panel out, with every filter and exclusion logged. If you cannot regenerate your dataset from scratch on week eleven, your results are not real.
Week five and six: exploration
Before any model, you look at the data. Distributions, correlations, time-of-day effects, symbol-level heterogeneity. You discover that the raw signal is dominated by a handful of large names on earnings days, which is not a persistent effect but will fool a naive backtest. You add a filter and a note explaining why.
You produce a dozen plots. Most of them are boring. A few of them tell you that the effect, if it exists, is small, concentrated in mid-cap names, and only visible after controlling for the overnight return of the broader index. This is useful. A null result with a clean explanation is a result.
Week seven to nine: model and backtest
You fit a simple model first, because a simple model you understand beats a complex model you do not. Linear regression, with the controls your exploration suggested, cross-validated by time. You document the specification choices honestly, including the ones that did not work.
Then the backtest. This is where small firms are strict and where students most often make mistakes. No lookahead. No survivorship bias. Realistic transaction costs, including spread and market impact estimates. A proper out-of-sample window. If the signal only works before costs, it does not work. You will be asked to explain, line by line, how a trade gets from signal to filled order, and the honest answer is often that it does not survive the trip.
Week ten to twelve: the writeup
The last three weeks are the writeup, and this is the part students most often underestimate. A ten-page document that another researcher could read and reproduce. A summary of the question, the data, the choices you made, the results, and the honest limitations. The limitations section is the one your supervisor reads most carefully, because it is the one that distinguishes someone who understands their own work from someone who does not.
You present it to the desk. You get questioned for an hour. You take notes. If the project moves forward in some form, it will usually be by somebody else; your contribution is a clean foundation, not a finished product. That is the normal and useful outcome. If the project is binned, that is also a result, and you move on without ego.
What you take with you
At the end, you will have one piece of work you can describe in detail: the data pitfalls, the specification choices, the controls, the costs, the limitations. That one project, told honestly, is worth more in future interviews than a resume full of vague claims. It is also the single best signal to the firm about whether to extend an offer.
If this is the kind of twelve weeks you want, our Quantitative Finance Internship is open. Applications go through a free assessment first; the internship itself follows the shape above.