IMC Prop Trading: Turn a Backtest into a Reproducible Research Note
IMC’s public research and careers page describes trading with its own capital and technology, with machine learning embedded in its research workflow. This article, checked 24 October 2026, explains that context and gives a small research-note exercise. It does not reproduce an IMC strategy or interview test.
Continue with this site’s separate research worksheet
What the source establishes about IMC
The first-party page links research, algorithms, systems and trading, and describes collaboration among researchers, engineers and traders. It also distinguishes model-focused and market-focused research roles. Those descriptions explain a professional trading business; they do not establish that a reader can buy an IMC evaluation account.
If the search is about a career, use the official careers information. The exercise below is independent educational work, not an assessment supplied, endorsed or scored by IMC. No personal role or Singapore eligibility was checked.
Keep accuracy and economic results in different columns
Imagine a fictional test of 100 signals with 60 correct direction labels. Directional accuracy is 60%. Now suppose the same toy experiment records 60 units of gross result and 80 units of all-in costs. Its net result is minus 20 units. The direction label alone would hide that outcome.
The example deliberately uses invented units and outcomes. It is not a forecast, a backtest report from IMC, a claim about an achievable strategy or evidence that every correct prediction produces equal profit. The point is to retain the definitions rather than optimise the headline.
| Quantity | Toy value | Meaning |
|---|---|---|
| Correct direction labels | 60 out of 100 | 60% label accuracy |
| Gross result | 60 invented units | Before assumed costs |
| All-in costs | 80 invented units | Separate model input |
| Net result | 60 − 80 = −20 units | Not inferred from accuracy alone |
A research note another person can inspect
Record the hypothesis before looking at the result; define the instrument identifiers, observation window, feature cutoff and outcome label. Keep the data used for choosing a model separate from the data reserved for its final check. Write down the cost assumptions alongside the gross calculation.
Then mark what has not been tested: whether the prices were executable, how latency and order handling would affect the result, and whether a different period changes the conclusion. A Singapore journal should retain the timezone of each observation and its conversion, rather than silently joining data by a local date.
Career research and programme selection have different evidence
An independent note can help the reader understand their own experiment; it does not certify job readiness, a hiring outcome or a provider reward. Preserve that boundary when comparing professional firms with retail programmes.
The HashHedge link below is optional for a separate evaluation-programme decision. It is not an IMC application, strategy licence or employment offer. Review HashHedge’s current conditions before registering.
Sources and rule dates
- IMC: AI/ML research, business and careers FAQ
Checked 2026-10-24. First-party own-capital, research and role descriptions; not a tested strategy, current vacancy guarantee or personal eligibility.