Middle East Technical University (ODTÜ) · Ankara, Turkey
BSc Physics · Penultimate year 2026–27, on track to graduate June 2028
Eren Ege Çelik
İzmir, Turkey · erenege3500@gmail.com · github.com/ErenEgeCelik
Independent quantitative researcher and prediction-markets trader. I formalize market mechanics as finite-state decision problems, derive expected value from observable transitions, and test policies with placebos, chronological out-of-sample evaluation and clustered uncertainty. Built Bayesian fair-value, latency-aware CLOB execution and microstructure research systems; traded small self-funded capital on Polymarket since 2026.
Education
İzmir Institute of Technology (İYTE) · İzmir, Turkey
Physics · GPA 3.6 / 4.0 · transferred to ODTÜ
İzmir Atatürk High School · İzmir, Turkey
Entered by national examination, top 0.7% nationally
Honors & selection
- Physics Olympiad Team, İzmir Atatürk High School — final year
- İzmir Mathematics Olympiad Team — selected in grade 8, the candidate pool for the national team
- İzmir Atatürk High School is among Turkey's five highest-scoring high schools by entrance rank
Experience
Independent Quantitative Researcher & Trader · Remote
Prediction Markets · Self-directed, live trading on Polymarket
- Built and operated a 24/7 research and execution stack on small self-funded capital, spanning probabilistic fair value, event-driven triggers, CLOB execution, replay, and live and paper verification.
- Grew a self-funded Polymarket account from roughly $30 to roughly $1,200 over three months of live trading, with no further deposits.
Weather fair value · Bayesian state space, Monte Carlo, calibration
- Built a fair-value engine for daily-temperature markets across ~28 cities: a scalar Kalman posterior over the true airport temperature, a Monte Carlo daily-maximum distribution, and a production-checked consistency identity linking the current, next-observation and conditional outcome distributions.
- Rebuilt the leakage-aware calibration pipeline; in offline replay it reduced mean Brier score by 24.2% against the prior model configuration across 28 market fits, with no fit regressing. The final calibration was never deployed.
Latency-aware execution and model retirement · Pre-signing, caching, replay
- Built a pre-signed EIP-712 order path with warm CLOB sessions, cached market state, keep-alive and multi-VPS routing; cut measured trigger-to-order latency from 10.2 s to ~58 ms, and measured one national feed running 4 min 44 s ahead of the reference source.
- Defined state, actions, transitions and reward explicitly; compared hold-to-resolution against exit-at-next-event policies under mean-variance utility and liquidity-aware exit assumptions.
- Retired the weather strategy after an out-of-sample replay over 3.3 GB, 50 city-days and 32 settlement events found −$0.58 expected value per $1 across every entry band. The median repricing window had compressed from 23 s to 0.15 s.
BTC 5-minute market microstructure · Brownian-probit, verifier-first research
- Reverse-engineered the dominant maker's Brownian-probit quote schedule to a ~0.92 median within-slot R² and ~6-tick chronological out-of-sample quote RMSE. The market's own prices calibrated outcomes better, so the model is structural replication rather than alpha.
- Defined an MDP-style state/action/transition tree and tested policies with placebo controls, chronological out-of-sample splits, slot-cluster bootstrap and feed-dropout cleaning. Fresh out-of-sample data rejected the static maker front at −0.98 cents per eligible quote moment, 90% CI [−1.63, −0.36].
- Measured simulator coverage directly: the same six crash slots replayed at +$21 while the contemporaneous paper arm lost $31. Positive tape results were treated as upper bounds and live-probe targets, never as deployment evidence.
World Cup cross-market relative value · Conditional goal model
- Built a conditional goal-arrival model and a precomputed trigger table; verified it against full simulation to a maximum divergence of 0.0028 and against live prices to within 0.02 before sizing small real-money positions.
Undergraduate Research Assistant (volunteer) · Ankara, Turkey
Prof. Ali Bozbey's Group, TOBB ETÜ · Superconducting / quantum-computing hardware
- Joined the group (which built Turkey's first superconducting quantum-computing hardware) on a volunteer basis after connecting with the team at ICSM; contributed to lab work over two weeks.
Engineering Intern · Stuttgart, Germany
Ingenieurbüro Bickele & Bühler GmbH · Electronics manufacturing & software
- Worked across SMD pick-and-place, AOI, laser and soldering operations, incoming-goods QC, and device assembly; wrote software in C# and Arduino.
Research & writing
Reverse-Engineering a Prediction Market Maker
Feed identification, volatility dynamics, and out-of-sample replication evidence from Polymarket BTC contracts.
Binary Market Making · in preparation
Inventory-skew derivation under CARA utility, adverse-selection analysis, and empirical verdicts under realistic execution.
Skills
- Programming
- Python (NumPy, pandas, asyncio, matplotlib), Linux, Git, Playwright, C# (2023)
- Methods
- Bayesian state-space models, MDPs, Monte Carlo, ordered probit, Brier calibration, chronological out-of-sample evaluation, placebo tests, clustered bootstrap
- Markets
- CLOB mechanics, order-book microstructure, Polymarket, Polygon and USDC execution, EIP-712 order signing, WebSocket and REST feeds
- Infrastructure
- AWS, multi-region deployment, VPS operations, Tailscale mesh
Activities
- Organizing Staff — International Conference on Superconductivity and Magnetism (ICSM) & ICSQMT, 2025 & 2026
Interests
Football — Göztepe academy U14–U15, school team captain in middle and high school. Competitive Valorant (Immortal, team tournaments). Chess.