Hey, I'm Srijan. I'm a maths student at SVNIT Surat, and pretty much all my interest lies in quant and sports analytics.

Right now I'm Head of Quant Engineering at Noether Meridian. A big part of the job is being the person who asks "are we sure?" Is the pricing code actually right? Does the backtest still look good once you count all the ideas that didn't work? Can someone else rerun it and get the same numbers?

When the honest answer is "not sure", that's the part I enjoy. I shrink the problem down to the smallest case that breaks, check it against something that has to be true however the code is written (a closed-form price, put–call parity, a curve that should reprice its own quotes), track down the cause, and leave a regression test behind so it can't quietly come back. I mostly write Python, plus Rust, C++ and R.

Kevin De Bruyne is the reason I follow football, and football is the reason PremierLock exists: an end-to-end, one-stop Premier League engine I'm building on the side.

I'm also starting a repository of notes on the maths I'm reading, partly so people can go through some simple maths (good luck 🙂) and partly to keep my LaTeX practice consistent. And I've finally started LeetCode. Again. 😕

Papers

A. K. Touré, S. Mishra, D. Shah · Preprint, Sep 2026

If you search a day of prices for the best-looking break, you will find one even when nothing happened. We fixed the test windows in advance, used block-bootstrap nulls, and controlled error across all 24 candidates with Romano–Wolf / max-T. In our Monte Carlo stress test (AR(1), intraday seasonality, Student-t shocks) the familywise size came out at 3.98% for a 5% target.

A. K. Touré, S. Mishra · Release paper for ASRQuant 1.3.0, Sep 2026

The paper for ASRQuant 1.3.0. I did the validation and benchmarking that its results and limitations sections are based on.

Work

Noether Meridian (founding member, Non-Executive Director)

2026 – now

Head of Quant Engineering, Alpha Stochastic Research (the firm's research lab) · remote

  • I lead the independent validation and benchmarking of ASRQuant 1.3.0 (PyPI), the lab's open-source Python library for quant research. It has 84 modules and 258 public functions and classes, covering curves, credit, derivatives, portfolios and backtesting.
  • We ran the release as 33 tracked work packages: install checks, API mapping, independent tests, numerical stability, cross-checks against references, benchmarks, fixes, and a final pass/block decision.
  • I tested the numbers against things that must hold no matter how the code is written: put–call parity, implied-vol round trips, repricing market quotes from fitted curves, recovering hazard curves from CDS quotes, and covariance matrices staying positive semi-definite.
  • I reviewed the anti-overfitting tools (CPCV, probability of backtest overfitting, Reality Check, SPA), reproduced and triaged bugs, and wrote regression tests for them.
  • The release shipped with 336 passing tests across 36 modules, all 84 modules importing cleanly, and a passing 116-symbol stable API contract.
  • I also work on QForge, the firm's proprietary C++20 engine for turning validated research into strategies. It takes an agreed strategy spec, implements it, runs it through deterministic replay and simulation with risk constraints, and produces evidence that can be reviewed independently before anything is delivered.

What I mean by alpha

Alice just rides z z z Bob rides and walks! ALPHA! the escalator lifts everyone = the market (beta) ↑ both got on here, at the same time
The escalator. Alice and Bob get on at the same time. Alice just stands there. Bob walks up the steps too. The escalator carries both of them up, and that part is the market, which everyone gets for free. The extra height Bob climbed on his own is alpha.
SO HOT! ALICE’S LEMONADE BOB’S LEMONADE sunny day: everyone sells more same sun + a tastier recipe 10 cups 10 cups + 3 extra = ALPHA
The lemonade stands. On a hot day every stand sells more lemonade. That's the weather, not skill. Bob sells three more cups than Alice on the same hot day. Those three cups are his alpha. The first ten were the sun.
100 kids each flip a coin 7 times… …plus 68 more kids one of them H 7 HEADS!! I’m a coin genius (nope, just luck: with 100 kids, someone gets 7 heads 54% of the time)
The lucky coin kid. Get 100 kids to flip a coin seven times and there's a better than even chance one of them gets seven heads. That kid isn't special. Test 100 trading ideas and the best one will look amazing for the same reason. Real alpha has to beat the luckiest kid, not zero.
Monday only you know the secret spot Friday everyone found it one fish, shared six ways lots of fish, all yours
The secret fishing spot. A great spot stops being great once everyone finds it. Same with alpha: once other traders copy an idea, there's less of it for each person. And what you take home is what's left after paying for the boat, so alpha only counts after costs.

If you're a quant

r − r_f = α + β·(factors) + ε

  • Escalator: β is the ride, α is the walking. Alpha is the intercept of a factor regression, so it only exists relative to a model. Add size, value and momentum and some of the walking turns out to be a faster escalator.
  • Lemonade: compare with the right benchmark on the same day, never with zero.
  • Coin kids: the best of N backtests is biased upward by selection. Deflated Sharpe ratio, a t > 3 hurdle, or a max-T bootstrap fix this. My regime-break paper is built on it.
  • Noise: real alpha is still hard to see. With an information ratio of 0.5 you need about 16 years of returns to get t = 2.
  • Fishing spot: alpha decays as it gets crowded and shrinks as you trade bigger. Only alpha after spread, impact and fees counts.

Projects

PremierLock

in progress

Python, pandas, scikit-learn, XGBoost, PuLP

A Premier League forecasting engine. A Dixon–Coles goal model and Monte Carlo season simulation give title, top-four and relegation odds. It also compares goals and points with xG and xPts to spot teams due to regress, forecasts transfer values, and has an FPL optimiser written as a mixed-integer program. Forecasts are backtested walk-forward against bookmaker odds.

Efficiency–Equity Nexus

Jan 2026

R, data.table, quantmod

Ranks India's top 20 banks on DEA efficiency (CCR and BCC), growth across regimes and price momentum, using RBI statements and Yahoo Finance data. The ranking was stress-tested with 5,000 Monte Carlo weight draws and turned into a portfolio with Buy, Accumulate, Hold and Avoid calls.

Mean–variance instability

Python, NumPy

Ten Nifty 50 stocks, two non-overlapping windows (2019–21 and 2022–24), and the closed-form Markowitz solution with no solver. Same stocks, same maths, very different weights. The estimation error in the means does most of the damage.

Education

SVNIT Surat

2024 – 2029

Integrated M.Sc. (5 years), Mathematics

IISER Pune, Summer School on Mathematical Finance

May – Jul 2026

With Dr. Anindya Goswami. Price processes, derivative pricing and portfolio optimisation. My certificate mentions my "critical investigation of the optimal portfolio solutions".

Certifications

Skills

Languages
Python, Rust, C++, R
Python
NumPy, pandas, SciPy, SymPy, scikit-learn, XGBoost, PuLP, Matplotlib
Quant
Stochastic processes, option pricing and implied vol, yield curves, credit (hazard rates, CDS), portfolio optimisation, bootstrap and multiple testing, structural breaks, Monte Carlo, backtest validation, LP/MILP
Tools
Git, Linux, Jupyter, VS Code

Other

  • Secretary & Chief Executive, Cutting Edge Visionaries (CEV), SVNIT
  • Finance & Documentation Co-Head at MindBend, Admin & Documentation Co-Head at Sparsh (SVNIT fests)

Contact me

Email is the best way to reach me. Happy to talk about quant research and engineering roles, validation work, the papers, or football.

srijanmishra070306@gmail.com

Talent is something you make bloom, instinct is something you polish.
· Ego Jinpachi, Blue Lock