Wrestling the World Into Rows with Eric Mannes

Career Trajectory & Motivation

• Eric Manis joined Jane Street after a decade-long tenure spanning trading, technology, and current leadership of the alternative data (alt data) effort.
• He selected Jane Street over academic research (Boston), tech (San Francisco), and other finance roles (New York) following three undergraduate summers, citing the trading environment as the most interesting fit.
• His trading internship revealed that finance is an "anti-inductive environment" where discovered patterns are competed away, forcing continuous discovery of new signals.
• The firm views the market as an information discovery system where participants are incentivized to learn but must share information, leading to market prices that act as reliable gauges of real-world value.

Trading Education & Simulation

• Jane Street invests significant resources in internship training, recognizing that few hires have prior trading experience.
• Interns engage in "mock trading" using low-stakes simulations, such as a trading card game called "Figgy," to learn decision-making under uncertainty without real financial pressure.
• These simulations include adversarial components where players trade based on private information, mirroring market dynamics where price discovery relies on resolving information asymmetries.
• Post-internship, traders are assigned to desks (e.g., commodities) with mentors to analyze real trades, focusing on why specific decisions were made and how to optimize behavior.
• Traders can influence firm strategy quickly by becoming "local experts" on niche instruments, eventually scaling from small suggestions to independent decision-making.

Commodities & Market Mechanics

• Manis served as a trading expert on the commodities desk, specializing in oil, natural gas, and index ETFs, requiring deep knowledge of physical market mechanics.
• He highlighted the complexity of futures contracts, including multipliers (e.g., 1,000 barrels for crude oil), margin requirements, and varying sign conventions for spreads across different exchanges.
• A critical "metadata problem" involves tracking different quoting conventions (dollars vs. cents) and contract specs to prevent massive trading errors.
• On April 20, 2020, the crude oil futures price settled at a negative value on the CME due to storage saturation in Cushing, Oklahoma, violating standard market assumptions.
• Negative pricing demonstrated that trading success requires engagement with "grotty details" of the physical world, such as storage capacity and supply/demand imbalances, not just mathematical modeling.
• The event caused failures in trading systems and risk models that assumed prices must always be positive, highlighting the fragility of assumptions in automated systems.

Technology Evolution & Desk Devs

• Jane Street transitioned from centralized infrastructure teams to a "desk dev" model, where each trading desk has embedded software engineers to build custom tools.
• Manis led the migration of the commodities risk system from slow, unmaintainable Microsoft Excel spreadsheets to a robust, type-safe software system.
• The Excel replacement addressed issues with version control, debugging, performance bottlenecks during critical moments, and the accumulation of undocumented logic ("spreadsheet artifacts").
• This transition required bridging the gap between domain experts (traders) who understood the risk but not code, and software engineers who needed deep domain context to build effective systems.
• The firm now employs embedded developers across all desks, reflecting a shift where traders must understand that investing in tooling reduces long-term operational pain and unlocks new strategies.

Alternative Data (Alt Data) Strategy

• Jane Street's alt data team was established to solve the problem of messy, hard-to-use external data that previously prevented the discovery of tradable signals.
• The team grew from zero data engineers in 2023 to over 20 in the current period, driven by "induced demand" as demonstrated value from early wins.
• The team is structured into three functional areas: Data Strategy (sourcing and vendor matchmaking), Data Infrastructure (building ingestion/processing tools), and Data Engineering (transforming raw data into usable datasets).
• Sourcing involves attending industry conferences (e.g., via data catalogs) to evaluate vendors offering satellite imagery, credit card transaction data, and other non-traditional datasets.
• The firm distinguishes its use of data from advertising: Jane Street seeks aggregated, anonymized macro trends rather than individual consumer targeting.

Data Engineering Challenges & Solutions

• Data engineers must solve complex domain problems, such as parsing unstructured financial reports (PDFs) into tabular data while accounting for corporate structures like dual-listed entities.
• A critical focus is "point-in-time" accuracy; engineers must ensure historical studies do not accidentally use corrected data that would not have been available at the time of a simulated trade.
• The firm maintains its own timestamps of data receipt to guard against vendor data corrections that could "smuggle" future information into the past.
• Tools like dbt (data build tool) are adopted for testing and composing transformations, while the firm retains a custom-built internal data warehouse ("Superstore") to leverage on-prem hardware for cost and latency efficiency.
• On-prem infrastructure allows for lower marginal costs compared to cloud models, avoiding the need to optimize queries solely for cloud billing units, though a hybrid approach is increasingly used.

Impact of AI and Machine Learning

• AI has increased the value of high-quality alternative data by enabling neural networks and LLMs to extract features from unstructured text more efficiently.
• LLMs present a "causality problem" where their training data may contain future knowledge, risking data leakage in backtesting if not carefully timestamped.
• While models can handle more raw data, the value of rigorous data cleaning and validation has increased; poor data quality leads to stronger models identifying non-tradable patterns more effectively.
• The verification bottleneck has become more critical, requiring human judgment to distinguish between genuine signals and AI artifacts, particularly regarding data provenance and vendor corrections.

Hiring & Talent Profile

• The firm seeks data engineers with curiosity, investigative rigor, and the ability to handle messy, unfamiliar real-world data.
• Ideal candidates often have backgrounds in science or social sciences where they have learned to make sense of noisy data, combined with strong software engineering skills.
• Interviews focus on "data investigation" scenarios where candidates must analyze an unfamiliar dataset, model its behavior, and document assumptions, testing for detail-orientation over raw coding speed.
• The team plans to introduce a paid internship program in 2025 to prepare candidates for full-time roles starting in 2029, emphasizing long-term talent pipeline development.