Quant analyst roles in India: research, risk and backtest hygiene

Four jobs that share the title

A model validator and a signal researcher may both hold the title "quant analyst", but one spends the week writing a validation report and the other testing hundreds of hypotheses on market data. Ask which one you are interviewing for.

The artefact: catching look-ahead bias in a backtest

The most common way a backtest lies is by using information that would not have been available at the time of the trade. This self-contained script uses synthetic prices with no real signal in them, then compares a correctly lagged strategy with one that accidentally peeks at the same day's return.

import random
import statistics

random.seed(42)
# Synthetic daily returns with no predictable pattern
returns = [random.gauss(0.0003, 0.012) for _ in range(1500)]

def strategy_returns(rets, lagged=True):
    out = []
    for t in range(1, len(rets)):
        # Signal: go long if the previous day's return was positive
        signal_source = rets[t - 1] if lagged else rets[t]   # the bug: using day t to trade day t
        position = 1 if signal_source > 0 else 0
        out.append(position * rets[t])
    return out

def annualised_sharpe(daily):
    mean, sd = statistics.mean(daily), statistics.stdev(daily)
    return (mean / sd) * (252 ** 0.5) if sd else 0.0

correct = strategy_returns(returns, lagged=True)
peeking = strategy_returns(returns, lagged=False)
print(f"Correctly lagged Sharpe: {annualised_sharpe(correct):.2f}")
print(f"Look-ahead Sharpe:       {annualised_sharpe(peeking):.2f}")

The lagged version shows the unremarkable result you would expect from random data. The peeking version shows a spectacular Sharpe ratio, because it only holds on days it already knows were up. In real code the bug is rarely this obvious: it hides in a feature computed with a rolling window that includes today's close, in a dataset of today's index constituents applied to past years (survivorship bias), or in corporate-action adjustments made with future information. A useful habit is to run every new strategy on data like this, where the true answer is "no edge", and check that the backtester agrees.

What the week looks like in two of the variants

A model validation quant at a bank's Indian centre might spend a week reviewing a new interest-rate model: reading the developer's documentation, replicating key results independently in Python, testing behaviour at extreme inputs, checking whether assumptions hold for the portfolio it will be used on, and writing findings graded by severity. Deadlines are set by model approval committees.

A quant researcher at a trading firm might spend it testing a handful of signal ideas: cleaning data, running backtests with realistic transaction costs, checking results across time periods, and presenting to a senior researcher who will try hard to find the flaw. Most ideas fail; recording why is part of the job.

Out-of-sample results and model validation findings

Validators: quality and timeliness of reviews, issues found before approval, and audit or regulatory findings. Front-office strats: reliability of pricing and risk tools, and responsiveness to traders. Researchers: the out-of-sample performance of signals that reach production, which is the only measure that counts in the end. Quant developers: correctness, performance and reliability of systems.

Desk heads, risk managers and quant developers

Depending on variant: a head of model risk, a strats lead reporting to a desk, a senior portfolio manager or research head, or an engineering lead. Colleagues include traders, risk managers, technologists, and in banks, auditors and regulators' reviewers.

Confused with data scientist and research analyst

A data scientist in a bank may do some quant-like modelling, and titles overlap. A research analyst in a KPO usually does fundamental equity or industry research, reading filings and building financial models, which is different from mathematical modelling. Check whether the posting asks for stochastic calculus and time-series statistics, or for financial statement analysis.

Stochastic calculus, C++ and backtesting in postings

• "Derivatives pricing, Greeks, stochastic calculus" — strats or pricing quant.
• "Model validation, SR 11-7, model risk" — validation (SR 11-7 is US Federal Reserve guidance on model risk management, often referenced by global banks).
• "Alpha research, backtesting, market microstructure" — quant research.
• "C++, low latency, systems" — quant developer.
• "VaR, expected shortfall, stress testing" — market risk quant.

Reproduce an existing result before proposing a new one

Whatever the variant: reproduce an existing model or backtest from scratch and reconcile your result to the official one. That exercise teaches the data, conventions and pitfalls faster than anything else. Then take ownership of one small component.

Senior quant, strats lead or portfolio research

Senior quant, desk strat, head of validation, portfolio manager in research tracks, or quant technology lead. The Data Science programme builds the statistics, time-based validation and leakage discipline that quant work demands, though specialised derivatives mathematics is beyond its scope.