Mathematics careers: actuarial, quant and data roles

Mathematics (B.Sc) · Careers and applications · free study notes for exams, placements and interviews

In short

A mathematics degree leads to three well-defined, well-paid career tracks in India: actuarial work (pricing and reserving for insurance and pensions, licensed by the Institute of Actuaries of India under the Actuaries Act, 2006 and regulated by IRDAI), quantitative finance (derivative pricing, risk and algorithmic trading at banks, hedge funds and market-making desks), and data science / analytics (statistical modelling and machine learning across every industry). All three rest on the same core mathematics — probability, statistics, linear algebra, calculus, optimisation and numerical methods — differing mainly in the time horizon of the risk modelled and the regulator involved. This topic is examined directly in B.Sc/M.Sc papers on actuarial statistics, financial mathematics and operations research, and indirectly in every campus placement test (quantitative aptitude, probability puzzles, SQL, Python) and in professional-body syllabi (IAI CS/CM/CB/SP series, CFA, FRM, CQF). Knowing the exam structure, the mathematics each role actually uses, and the standard interview problems is the difference between a maths degree and a maths job.

Explained

The three tracks compared

Actuary - quantifies long-horizon contingent risk: mortality, morbidity, lapse, claims, interest. Horizon 1-60 years. Tools: life tables, compound interest, GLMs, run-off triangles, stochastic asset-liability models.
Quant - prices and hedges financial instruments and manages market risk. Horizon microseconds (HFT) to years (structured products). Tools: stochastic calculus, PDEs, Monte Carlo, time series, numerical linear algebra, C++/Python.
Data scientist / analyst - extracts decisions from data. Horizon days to months. Tools: regression, classification, clustering, causal inference, A/B tests, SQL, Python, cloud pipelines.
Degree fit: B.Sc Mathematics (Hons) with statistics is sufficient entry for all three; quant research roles usually want M.Sc/M.Stat/M.Tech or a PhD; actuarial progression is by examination, not degree.
Overlap: all three need probability distributions, estimation, linear algebra, optimisation and the ability to write correct, tested code. Only the actuary needs a statutory licence.

Actuarial track: structure, exams, regulation

Regulator and body: the profession is governed by the Actuaries Act, 2006; the Institute of Actuaries of India (IAI) sets exams and issues Fellowship. Insurers are regulated by IRDAI (IRDA Act, 1999), which requires every insurer to appoint an Appointed Actuary who certifies reserves and solvency.
Entry: clear the ACET (Actuarial Common Entrance Test) to become an IAI student member. Exemptions from some papers are available to graduates of certain recognised university programmes - check the IAI's current exemption list, as it changes.
Exam stages (IAI, aligned with the UK IFoA structure):
1. Core Principles - CS1 Actuarial Statistics, CS2 Risk Modelling and Survival Analysis, CM1 Actuarial Mathematics, CM2 Financial Engineering and Loss Reserving, CB1 Business Finance, CB2 Business Economics, CB3 Business Management.
2. Core Practices - CP1 Actuarial Practice, CP2 Modelling Practice, CP3 Communications Practice.
3. Specialist Principles (SP) - choose two: SP1 Health and Care, SP2 Life, SP4 Pensions, SP5 Investment, SP6 Financial Derivatives, SP7 General Insurance Reserving, SP8 GI Pricing, SP9 Enterprise Risk Management.
4. Specialist Advanced (SA) - one paper, plus work-experience requirements, for Fellowship (FIAI). Associateship is an intermediate qualification.
Fee and timing: exams are held twice a year (typically March/April and September/October sessions); the number of papers, exemption rules and fees are revised periodically, so verify on the IAI site before planning.
What the work is: pricing (setting premium rates), valuation/reserving (how much must be held for future claims), experience analysis (actual versus expected mortality/lapse), solvency and capital, reinsurance, embedded value, IFRS 17 / Ind AS 117 reporting.
Employers in India: LIC, SBI Life, HDFC Life, ICICI Prudential, Max Life, Bajaj Allianz, Tata AIA; general insurers ICICI Lombard, New India Assurance; reinsurer GIC Re; consultancies Milliman, WTW, Mercer, Aon, Deloitte, EY; GCC/offshore actuarial centres of Swiss Re, Munich Re, Prudential, Aviva, MetLife (Gurugram, Mumbai, Bengaluru, Pune).
Advantages: statutory scarcity (a few thousand Fellows for a large market), high pay per exam cleared, strong job security. Limitations: 6-10 years to qualify, low pass rates per sitting (commonly 20-40%), self-study alongside full-time work.

Quantitative finance track

Sub-roles:
- Quant researcher / strategist - builds alpha signals and pricing models.
- Quant developer - implements models in production C++/Python; lowest-latency work at market makers.
- Quant risk / model validation - independently tests models; heavy demand in bank GCCs under RBI and Basel norms.
- Quant trader - executes and manages a book using the models.
Mathematics used: probability and measure-theoretic basics, Itô calculus and SDEs, martingales and risk-neutral pricing, PDEs (Black-Scholes-Merton equation, finite differences), Monte Carlo with variance reduction, numerical linear algebra (Cholesky, SVD, PCA for yield curves), optimisation (mean-variance, convex programming), time series (ARIMA, GARCH), statistical learning.
Core result to know: under no-arbitrage and a complete market, a derivative price is the discounted expectation of its payoff under the risk-neutral measure; the Black-Scholes formula is the closed-form answer for a European option on a lognormal underlying.
Indian market context: NSE and BSE derivatives (index and stock futures and options), currency and commodity derivatives (MCX), government bonds (NDS-OM), interest-rate swaps; regulated by SEBI (algorithmic trading approvals, co-location, order-to-trade ratios) and RBI (bank treasury, FX). Quant roles also sit in GIFT City (IFSCA-regulated) entities.
Employers: market makers and prop firms (Tower Research, Optiver, Quadeye, Graviton, AlphaGrep, WorldQuant, Da Vinci, Jump), bank GCCs (Goldman Sachs, Morgan Stanley, JPMorgan, Barclays, Deutsche, HSBC, Nomura, UBS), Indian institutions (ICICI, Axis, Kotak treasuries), mutual funds and NBFC risk teams.
Credentials that help: CFA (Levels I-III), FRM (GARP), CQF; strong competitive-programming or Olympiad record; a published project in time series or pricing. None is a substitute for a demonstrated probability/coding interview performance.

Data and analytics track

Role ladder: data analyst → data scientist → ML engineer / applied scientist → decision scientist; parallel ladders in business analytics, risk analytics and operations research.
Mathematics used: descriptive statistics, hypothesis testing, confidence intervals, regression (linear, logistic, Poisson, GLMs), regularisation (ridge, lasso), resampling (bootstrap, cross-validation), Bayesian inference, dimension reduction (PCA), clustering (k-means, hierarchical), tree ensembles, gradient-based optimisation, experiment design and causal inference (difference-in-differences, propensity scores, instrumental variables).
Engineering skills expected: SQL (joins, window functions, CTEs) is non-negotiable; Python with NumPy, pandas, scikit-learn, statsmodels; version control; one cloud (AWS/Azure/GCP); dashboarding (Power BI/Tableau).
Indian employers: Flipkart, Amazon, Swiggy, Zomato, Zepto, Meesho, PhonePe, Paytm, Razorpay, Jio, Airtel; banks and NBFCs (HDFC, ICICI, Bajaj Finance) for credit risk and collections scoring; analytics firms (Mu Sigma, ZS Associates, Fractal, Tiger Analytics, AbsolutData, Quantiphi); consultancies and GCCs (Accenture, Deloitte, EXL, WNS, Tredence); public sector analytics (MoSPI, NITI Aayog, RBI DEPR research posts, IIM/ISI research assistantships).
Compliance context: personal-data work in India is governed by the Digital Personal Data Protection Act, 2023 (consent, purpose limitation, data-principal rights); credit scoring is also constrained by RBI's fair-practice and CIC norms. Model explainability matters where an adverse decision is made about a person.
Advantages: fastest entry (open to fresh graduates), widest employer base, remote-friendly. Limitations: title inflation, large low-value reporting component at the junior end, skills churn.

Entrance routes, higher study and government exams

M.Sc/M.Stat/M.Tech entries: IIT JAM (M.Sc Mathematics / Mathematical Statistics at IITs, NITs, IISc), ISI Admission Test (B.Stat, M.Stat, MS in Quantitative Economics), CMI entrance (B.Sc/M.Sc Mathematics), TIFR GS, IIT JEE Advanced for the five-year integrated M.Sc at IIT Kanpur/Kharagpur/Bombay (via JAM for PG), CUET-PG for central universities.
Research and fellowships: CSIR-UGC NET (Mathematical Sciences) for JRF/Lectureship, GATE Mathematics (MA) for M.Tech/PSU/JRF, NBHM scholarships and PhD screening, INSPIRE for B.Sc/M.Sc.
Specialised PG in finance/analytics: M.Sc Quantitative Finance / Financial Engineering, ISI's MS in QE, IIM/IIT business-analytics programmes, IGIDR, MDI, IIQF/CQF short courses.
Government and teaching routes: RBI Grade B (DR - General and DEPR/DSIM streams; DSIM explicitly wants statistics/mathematics PG), SEBI Grade A, NABARD, ISS (Indian Statistical Service) through UPSC, IES (Indian Economic Service), SSC CGL statistical posts, NIC/C-DAC, DRDO/ISRO scientist posts in operations research and cryptology, CSIR labs, plus NET/SET for assistant-professor eligibility and KVS/NVS/state TGT-PGT for teaching.
Actuarial-adjacent: IRDAI's own actuarial and insurance posts; LIC AAO (Actuarial stream).

Placement process and how to prepare

Typical pipeline: CV shortlist → online test (aptitude + probability + puzzles, or SQL/Python for data roles) → technical interviews (2-4 rounds) → HR/fit round.
Online test content: probability (conditional, Bayes, expectation), combinatorics, number puzzles, mental arithmetic under time pressure, basic statistics, SQL queries, short coding problems (arrays, strings, dictionaries), market-making/betting games for trading roles.
Must-build portfolio: two or three end-to-end projects with the data source, assumptions, model, validation metric and limitation stated. For actuarial, a mortality-table or claims-triangle project; for quant, an options backtest or GARCH volatility forecast; for data, a classification model with cross-validated metrics.
Compensation (approximate Indian ranges; market-dependent and changing fast): actuarial trainee ₹5-9 lakh per annum rising sharply with exams cleared; qualified actuary commonly ₹25 lakh and above; quant analyst at a top prop firm ₹30 lakh to over ₹1 crore total compensation at entry; bank quant/risk ₹10-25 lakh; data analyst ₹5-10 lakh; data scientist ₹12-30 lakh. Treat these as orders of magnitude, not quotes.
Tax note: salary is taxed under the Income-tax Act, 2025 (in force from 1 April 2026), which replaced the Income-tax Act, 1961 and renumbered its sections and forms; standard deduction and the default-regime slabs apply to salaried offers, and TDS is deducted monthly by the employer. Joining bonuses and stipends are taxable; a clawback repaid later does not automatically reverse the tax already deducted.
Employment terms: the four Labour Codes (Code on Wages; Industrial Relations; Social Security; Occupational Safety, Health and Working Conditions) are in force from 21 November 2025 and replace the earlier Factories, Minimum Wages, Industrial Disputes and BOCW Acts - they govern the appointment letter, wage definition for PF/gratuity and fixed-term employment.

Key points

Actuary - licensed professional who prices and reserves long-horizon contingent risk; qualification is by IAI examination, not by degree.
Actuaries Act, 2006 - statute creating the IAI and the protected title "Actuary"; IRDAI mandates an Appointed Actuary for each insurer.
ACET - IAI entrance test for student membership; the exam ladder is CS/CM/CB → CP → SP (two) → SA for Fellowship.
Reserving - estimating liabilities for claims already incurred; pricing - setting premiums for future risk. Different techniques, same data.
Quant - prices and hedges derivatives using stochastic calculus and numerics; splits into research, developer, risk and trading sub-roles.
Risk-neutral pricing - price = discounted expected payoff under a measure where assets drift at the risk-free rate; the basis of Black-Scholes.
Greeks - sensitivities of an option price: delta (to spot), gamma (rate of change of delta per unit spot, i.e. the second derivative in spot), vega (to volatility), theta (to time), rho (to rate).
Data scientist - builds predictive and causal models; SQL plus Python plus a validated metric is the entry bar.
GLM - generalised linear model with a link function; the industry standard for motor and health insurance pricing and for credit scoring.
Monte Carlo - price or risk by simulating many paths; error falls as 1/√N, so 100× more paths gives 10× accuracy.
VaR and expected shortfall - loss not exceeded at a confidence level, and the average loss beyond it; ES is the Basel-preferred measure.
Life table - lx, dx, qx, px columns from which survival probabilities and annuity values are built; India uses IALM-type mortality tables for assured lives.
Credentials - CFA, FRM, CQF signal intent in finance; IIT JAM, ISI, CMI, GATE MA, CSIR-NET, RBI Grade B (DSIM) are the formal Indian routes.
DPDP Act, 2023 - governs personal-data processing in analytics work: consent, purpose limitation, data-principal rights.
SEBI and RBI - regulate algorithmic trading and bank treasury/model risk respectively; compliance shapes what a quant may deploy.
Common core - probability, statistics, linear algebra, optimisation and tested code are required in all three tracks; only the horizon and regulator change.
Denominator discipline - every reported model metric must carry its sample size and validation split; an unqualified accuracy figure is not a result.

Formulas and examples

Compound interest / accumulation: A = P(1 + i)ⁿ ; v = 1/(1 + i) is the discount factor. P in ₹, i per period (decimal), n periods.
Effective rate from nominal compounded m times a year: i = (1 + i⁽ᵐ⁾/m)^m − 1. Force of interest δ = ln(1 + i).
Annuity-immediate present value: a_n = (1 − vⁿ)/i ; annuity-due: ä_n = a_n × (1 + i). Payment of ₹1 per period.
Life-table probabilities: p_x = l_{x+1}/l_x ; q_x = 1 − p_x ; _tp_x = l_{x+t}/l_x. l_x = number alive at age x.
Expected present value of a whole-life assurance (discrete): A_x = Σ_{k=0}^{∞} v^{k+1} · _kp_x · q_{x+k}, benefit ₹1 payable at end of year of death.
Net level annual premium (whole life, benefit S): P = S · A_x / ä_x — equivalence principle: EPV(premiums) = EPV(benefits).
Black-Scholes European call: C = S₀N(d₁) − K e^{−rT} N(d₂), with d₁ = [ln(S₀/K) + (r + σ²/2)T] / (σ√T), d₂ = d₁ − σ√T. S₀, K in ₹; r, σ per annum; T in years; N(·) standard normal CDF.
Put-call parity: C − P = S₀ − K e^{−rT}. Delta of a call = N(d₁).
Parametric VaR (one period): VaR_α = −(μ + z_α σ) × V, z_{0.01} = −2.326, z_{0.05} = −1.645; V = portfolio value in ₹.
Sharpe ratio = (R_p − R_f)/σ_p ; annualise a daily Sharpe by × √252.
OLS estimator: β̂ = (XᵗX)^{−1} Xᵗy ; logistic model: ln[p/(1 − p)] = Xβ, so p = 1/(1 + e^{−Xβ}).
Monte Carlo standard error = s/√N ; chain-ladder development factor f_j = Σ C_{i,j+1} / Σ C_{i,j} over overlapping rows.
Classification metrics: Precision = TP/(TP+FP), Recall = TP/(TP+FN), F1 = 2PR/(P+R), Accuracy = (TP+TN)/Total.

Example 1 - Annuity and premium (actuarial, CM1 style). A 3-year term annuity-immediate of ₹1,00,000 a year at i = 8%.

v = 1/1.08 = 0.925926; v² = 0.857339; v³ = 0.793832.
a₃ = (1 − 0.793832)/0.08 = 0.206168/0.08 = 2.57710.
PV = 1,00,000 × 2.57710 = ₹2,57,710.
Annuity-due: ä₃ = 2.57710 × 1.08 = 2.78327 → PV = ₹2,78,327 (one year less discounting).

Example 2 - Life table and a one-year premium. l₄₀ = 95,000, l₄₁ = 94,620, i = 6%, sum assured ₹10,00,000, benefit at end of year.

q₄₀ = (95,000 − 94,620)/95,000 = 380/95,000 = 0.004 (4 per 1,000).
EPV of benefit = 10,00,000 × 0.004 × (1/1.06) = 4,000 × 0.943396 = ₹3,773.58.
The net single premium is ₹3,773.58; the office premium adds expenses, commission, and a margin for adverse deviation; there is no GST to add, because premiums on individual life (and health) policies have been GST-exempt since 22 September 2025 (earlier 18% on the risk premium).

Example 3 - Black-Scholes call (quant interview standard). S₀ = ₹100, K = ₹100, r = 10% p.a., σ = 20% p.a., T = 1 year.

d₁ = [ln(1) + (0.10 + 0.02)×1]/(0.20×1) = 0.12/0.20 = 0.60; d₂ = 0.60 − 0.20 = 0.40.
N(0.60) = 0.7257, N(0.40) = 0.6554.
Ke^{−rT} = 100 × e^{−0.10} = 100 × 0.904837 = 90.4837.
C = 100 × 0.7257 − 90.4837 × 0.6554 = 72.57 − 59.30 = ₹13.27.
Put by parity: P = C − S₀ + Ke^{−rT} = 13.27 − 100 + 90.48 = ₹3.75. Call delta = N(d₁) = 0.726, so hedge one short call with 726 shares per 1,000 options.

Example 4 - VaR and Sharpe (risk role). Portfolio ₹50 crore, daily mean return 0.04%, daily σ 1.2%, risk-free 6% p.a.

1-day 99% VaR = −(0.0004 + (−2.326)(0.012)) × 50,00,00,000 = −(0.0004 − 0.027912) × 50 cr = 0.027512 × 50 cr = ₹1.3756 crore.
10-day VaR (square-root-of-time) = 1.3756 × √10 = 1.3756 × 3.1623 = ₹4.350 crore.
Annualised return ≈ 0.0004 × 252 = 10.08%; annualised σ = 0.012 × √252 = 0.012 × 15.8745 = 19.05%.
Sharpe = (10.08 − 6)/19.05 = 0.214 — thin; a prop desk would want well above 1.

Example 5 - Chain ladder and a model metric (general insurance plus data). Cumulative paid claims (₹ lakh):

Accident year	Dev 1	Dev 2	Dev 3
2023	100	150	180
2024	120	174	—
2025	140	—	—

f₁ = (150 + 174)/(100 + 120) = 324/220 = 1.47273; f₂ = 180/150 = 1.20.
2024 ultimate = 174 × 1.20 = 208.8; 2025 ultimate = 140 × 1.47273 × 1.20 = 206.182 × 1.20 = 247.42.
Reserve = (208.8 − 174) + (247.42 − 140) = 34.8 + 107.42 = ₹142.22 lakh.
Now a fraud model on 10,000 claims with 200 frauds flags 150, of which 90 are genuine frauds: Precision = 90/150 = 0.60, Recall = 90/200 = 0.45, F1 = 2(0.60)(0.45)/(1.05) = 0.54/1.05 = 0.514. Accuracy = (90 + 9,740)/10,000 = 98.3% — high only because 98% of claims are not fraud, which is why accuracy is the wrong headline metric here.

Dimension	Actuarial	Quant finance	Data science
Qualification	IAI exams (ACET → CS/CM/CB → CP → SP → SA)	Degree + CFA/FRM/CQF optional	Degree + portfolio
Core maths	Compound interest, life contingencies, GLM, run-off triangles	Stochastic calculus, PDEs, Monte Carlo, time series	Regression, ML, causal inference
Primary code	R, Python, Excel, Prophet/MoSes	C++, Python, kdb+/q	Python, SQL, Spark
Regulator	IRDAI, PFRDA, IAI	SEBI, RBI, IFSCA	DPDP Act, 2023; sectoral norms
Horizon	Years to decades	Microseconds to years	Days to months
Entry speed	Slow (6-10 yrs to qualify)	Hard filter, fast pay	Fastest entry

Where it is used

Life insurance and pensions: product pricing, statutory valuation, embedded value and IFRS 17 / Ind AS 117 reporting at LIC, HDFC Life, SBI Life; NPS and superannuation work under PFRDA.
General and health insurance: motor and health GLM pricing, IBNR reserving via chain ladder and Bornhuetter-Ferguson, reinsurance structuring at ICICI Lombard, New India Assurance, Star Health, GIC Re.
Capital markets: options market-making and index arbitrage on NSE, volatility forecasting, execution algorithms at Quadeye, Graviton, AlphaGrep, Tower Research; derivative valuation and XVA at bank GCCs in Mumbai, Bengaluru and Gurugram.
Banking risk: PD/LGD/EAD credit models, IFRS 9 / Ind AS 109 expected credit loss, market-risk VaR and stress testing under RBI's Basel III implementation; model validation teams.
Consumer internet and fintech: demand forecasting, pricing and surge models, recommendation systems, fraud and credit underwriting at Flipkart, Swiggy, PhonePe, Razorpay, Bajaj Finance.
Public sector and research: MoSPI survey statistics, RBI DEPR/DSIM economic modelling, ISI and IIT research, operations research in Indian Railways scheduling, ISRO/DRDO cryptology and optimisation.

Interview questions

Q: What does an actuary do that a data scientist does not? A: An actuary signs off statutorily required reserves and solvency figures under the Actuaries Act, 2006 and IRDAI rules, carrying professional liability for a long-horizon liability estimate; a data scientist's model has no statutory certification.
Q: State the equivalence principle for premiums. A: At issue, the expected present value of future premiums equals the expected present value of future benefits and (for an office premium) expenses, valued using the chosen mortality and interest assumptions.
Q: A fair coin is tossed until the first head. What is the expected number of tosses? A: It is geometric with p = 0.5, so E[N] = 1/p = 2 tosses.
Q: Why is accuracy a poor metric for fraud detection? A: With a 2% fraud base rate, predicting "never fraud" scores 98% accuracy while catching nothing; precision, recall, F1 or AUC-PR measure the rare class properly.
Q: What does the Black-Scholes delta mean operationally? A: Delta is ∂C/∂S, the number of shares needed per option to make the position locally insensitive to spot; it is N(d₁) for a call and is rebalanced as spot, time and volatility move.
Q: Why does Monte Carlo accuracy improve only as 1/√N? A: The estimator is a sample mean of independent draws, whose standard error is σ/√N, so cutting the error in half needs four times the paths; variance reduction or quasi-random sequences beat brute force.
Q: Difference between a GLM and ordinary least squares? A: OLS assumes a normal response with identity link; a GLM allows exponential-family responses (Poisson for claim counts, gamma for severity, binomial for default) with a link function, fitted by maximum likelihood.
Q: What is the chain-ladder assumption, and when does it fail? A: It assumes future development factors mirror past ones for all accident years; it fails after a change in claim settlement speed, a law change, large claims, or when exposure mix shifts — then Bornhuetter-Ferguson or a frequency-severity model is used.
Q: VaR versus expected shortfall? A: VaR is the loss threshold not breached with probability α and ignores the shape of the tail beyond it; ES is the mean loss conditional on exceeding VaR, is sub-additive and hence coherent, and is the Basel-preferred measure.
Q: Two envelopes contain ₹X and ₹2X; you open one and see ₹100. Should you switch? A: Not on the naive "expected 125" argument — that assumes an improper uniform prior over X; with any proper prior the expected gain depends on the prior and is not uniformly positive.
Q: Which single skill do you most need for a data role you cannot learn from maths alone? A: SQL at production level — joins, window functions, CTEs and query cost — because the data must be correctly assembled before any model is justified.
Q: How would you check whether a trading strategy's backtest is real? A: Out-of-sample and walk-forward testing, transaction costs and slippage, capacity against traded volume, a count of parameters and trials (to penalise multiple testing), and stability of the Sharpe across sub-periods and regimes.

Common mistakes

Treating annuity-immediate and annuity-due as interchangeable — the due version is larger by a factor (1 + i).
Using nominal rates without converting to effective ones, or mixing per-annum σ with daily returns in the same formula.
Quoting a model metric with no sample size, class balance or validation split — an unsupported accuracy figure is not a result.
Confusing pricing with reserving in actuarial interviews: pricing looks forward at unwritten risk, reserving looks back at incurred claims.
Assuming the Black-Scholes assumptions hold — constant volatility, lognormal prices, no jumps, continuous hedging — and ignoring the volatility smile that proves they do not.
Scaling VaR across horizons with √t when returns are autocorrelated or the portfolio contains options with non-linear payoffs.
Planning a career around a compensation figure or an exam-exemption rule read once — both IAI exemptions and market pay change; verify before committing.