Skip to content
MINGYANG FUNDIndependent investment research
The researcher

Alex Chen

Fundamental investing, data and machine learning. B.S. Economics & Computer Science at Emory University.

Background

Alex Chen
Independent equity researcher · Emory University (B.S. Economics & Computer Science)
BNP Paribas — Global Markets Summer Analyst, CLO Structuring (2026) · Nomura Securities — Equity Research Summer Analyst (2025) · Manulife Investment Management — Investment Management Summer Analyst (2024)

I run a long-biased, fundamentals-first equity book across U.S. and China markets. The starting point is top-down macro; the work is bottom-up: unit economics, channel checks and primary data, with every thesis written down, dated and scored against what actually happened.

At Nomura Securities in Shanghai I was an Equity Research Summer Analyst on the consumer-staples team: I initiated coverage on the global tobacco sector with full financial and valuation models for eight listed companies, and co-authored the 20-page initiation report and roadshow deck presented with the institutional sales team to 30+ clients. At Manulife Investment Management in Shanghai I was an Investment Management Summer Analyst working with a fixed-income portfolio manager, and led bottom-up research on China's DRAM sector that contributed to a fund ranked first of roughly 3,000 China A-share equity funds in the first half of 2024. In summer 2026 I was a Global Markets Summer Analyst in CLO Structuring at BNP Paribas in New York. I have passed CFA Level I and am a Level II candidate.

At Emory I am a research assistant in Prof. Jin Wei's group, testing whether large-language-model reasoning adds value in trading beyond quantitative baselines. This site is the working record of my own process: live theses, a track record with its evidence, a forecast ledger, and post-mortems on what did not work. It is a personal account with no outside capital.

Experience & investment approach

Investment philosophy
  • Top-down macro first: macro sets the table, stock selection fills the plate
  • Concentrated book of high-conviction ideas, not diversification for its own sake
  • Find the value/growth discrepancy: quality growing faster than the market believes
  • Long-biased; selective, defined-risk shorts only when the risk/reward is asymmetric
  • U.S. and China focus, where the informational and analytical edge is real
  • Every thesis gets a date, a price, a target and a kill criterion
Experience & credentials
BNP Paribas — Global Markets Summer Analyst, CLO Structuring (2026)Nomura Securities — Equity Research Summer Analyst (2025)Manulife IM — Investment Management Summer Analyst (2024)Rand & Co Holdings — Private Equity InternCFA Level I passed · Level II candidateEmory University — B.S. Economics & Computer SciencePython · SQL · Bloomberg · FactSet · Capital IQEnglish · Mandarin
Areas of focus
U.S. EquitiesChina EquitiesBasic MaterialsEnergy InfrastructureConsumerTechnology / SaaSSpecial SituationsMacro-Driven Thematic

Research & code

Does LLM reasoning add value in trading?

Research assistant in Prof. Jin Wei's group at Emory (since Feb 2026). Built an end-to-end pipeline on 40–60 U.S. consumer-discretionary firms (2017–2025) using the FINSABER bias-mitigation framework, with a fine-tuned DeBERTa classifier, an elastic-net forecaster and a Kronos foundation-model variant, to test whether LLM reasoning adds alpha beyond quantitative baselines across market regimes. Leading a multi-agent LLM trading-system study with a two-layer evaluation protocol that separates reasoning quality from cascade effects, extending FinAgent, FINCON and TradingAgents.

What LLMs can and cannot do in my process

They parse earnings calls and filings, summarise news flow and extract covenant and pricing terms from documents (a pipeline of this kind became the standard scrubbing workflow on the CLO desk). They do not set probabilities, targets or sizing: every number on this site is a model or a judgement I can defend line by line, and the forecast ledger scores those judgements, not the tools.

The El Niño tracker's data pipeline

A serverless function pulls four public NOAA CPC files (weekly and monthly Niño3.4, ONI and relative ONI), a shared parser turns them into series for both the live page and the static build, delayed quotes come through a keyed proxy, and a validator refuses any deploy whose probabilities do not sum to one or whose forecasts are missing from the ledger. Site, renderers, functions and validation are my own code.

Contact

Explore the process

Follow an idea from variant view and primary evidence to valuation, catalysts and the post-mortem.

Investment process ↗