Trading & Capital Markets
Quantitative Researcher
The job is not finding signals. It is disproving them, quickly and honestly, because a signal you believed and should not have is far more expensive than one you missed.
The job is not finding signals. It is disproving them, quickly and honestly, because a signal you believed and should not have is far more expensive than one you missed.
A day that is actually representative
| Time | What you are actually doing |
|---|---|
| 08:00 | Read the overnight. Not for news, for whether anything invalidates a live assumption. |
| 09:15 | A hypothesis from yesterday failed its holdout. Write down why before moving on, because that note is the actual asset. |
| 11:00 | Data work. Corporate actions, survivorship, point-in-time correctness. This is most of the job and nobody warns you. |
| 14:00 | A research meeting where your result is attacked. The attack is the value; a result nobody attacked is not evidence. |
| 16:00 | Live monitoring — a strategy in production is decaying and you have to decide whether it is noise or regime. |
| 18:00 | Reading. Not papers necessarily; frequently exchange notices, because a rule change can end a strategy overnight. |
The decisive round
“You have a backtest showing a Sharpe of 2.4.”
Interviewer: “Talk me through whether you believe it.”
A weak answer. “A Sharpe of 2.4 is strong. I'd check it's robust across different periods, run it out of sample, and make sure transaction costs are included. If it holds up, I'd propose paper trading it.” Every candidate says this. It lists hygiene checks without saying what would make the result false, and it treats a high Sharpe as a claim to be defended rather than a red flag to be explained.
A strong answer. “My honest first reaction to 2.4 is suspicion, not excitement, and I'd say that out loud — because at any capacity worth having, a Sharpe of 2.4 is high enough that the most likely explanation is a bug in my own code.
So I'd go looking for the bug in a specific order.
Lookahead. The most common and the most seductive. Did I use a price that wasn't knowable at decision time? Did I use adjusted close, which encodes future splits and dividends into the past? Did I filter a universe using data that only exists today — every index constituent list is a lookahead trap unless it's point-in-time. Did my fundamentals arrive on the report date or the filing date? For India, did I use a corporate-action-adjusted series without checking the adjustment date?
Survivorship. If my universe is 'stocks currently in the NIFTY 500', I have quietly excluded everything that was delisted, merged or collapsed. That alone can manufacture a large part of a fake edge.
Multiple testing. How many variants did I try before this one? If I tested two hundred parameter sets, a Sharpe of 2.4 in the best one is roughly what randomness produces. The fix is to hold out data I have literally never looked at, and to record how many hypotheses I've run — a research log, so the denominator is honest.
Costs, properly. Not a flat basis-point assumption. For India: STT on the correct leg, exchange charges, stamp duty, GST on brokerage, and — the one that kills most equity strategies — realistic slippage and impact at my size. A strategy that trades the open with a market order does not get the open price.
Then the control I'd insist on. Run the identical machinery against a randomised version of the signal — same universe, same turnover, same holding period, shuffled entry. If the random version also produces a decent Sharpe, my edge is in the exposure and the trading pattern, not the signal. That control has killed more of my ideas than any of the checks above, and it's the one most people skip.
Only after all of that would I care whether it's out of sample.”
Treating a high Sharpe as evidence of a bug, naming lookahead precisely rather than generically, counting hypotheses, and — the differentiator — running a matched random control. That last move is what separates a researcher from someone who has run backtests.
What to build or bring
- A research log showing hypotheses you rejected, with why. This is the single most persuasive artefact in this field and almost nobody brings one.
- One study, end to end, on public data — with the control, the costs, and an honest conclusion. A well-executed null result is more impressive than a suspicious win.
- Probability and statistics you can do out loud. Conditional probability, Bayes, expectation, variance of a sum. It will be tested conversationally, not on paper.
- Know the market you are applying to. Indian F&O microstructure is not US equity microstructure, and answering with the wrong one is very visible.
Compensation
| Market | Band | Notes |
|---|---|---|
| United States | $250k – $600k+ total | Top prop firms pay new PhDs at the low end of that and scale fast |
| India | ₹40L – ₹1.5Cr+ | Prop firms (Quadeye, Graviton, AlphaGrep, Tower, Optiver, Jane Street) pay far above banks |
The book for this field
Technical Analysis
Twelve chapters that teach charts the way a researcher would: what each claim actually is, what the evidence supports, and the control you would need to believe it. Structure, levels, moving averages, oscillators, volatility, volume, candlestick patterns, Smart Money Concepts assessed honestly, and a full methodology for testing a rule without fooling yourself.
Cross-cutting
Skills every archetype tests
These are shared across every field on this site — the same question asked in different vocabulary — so preparation here compounds rather than being spent once. The book above is specific to your field.
Practice the 10 Trading questions → · Back to Trading & Capital Markets →