01. I started researching investments on my own
Small capital, simple rules, and the hope of building a research process I could run alone.
Read articleWriting
Notes and essays
Notes and essays by Donggyu Lee on machine learning, economics, computational social science, and research.
02. I asked an LLM to invent trading factors
Local Qwen, a growing factor bank, and what survived re-evaluation.
Read article03. Would buying a sharp crypto dip work?
How a simple EMA dip rule changed, and why its backtest was not a live track record.
Read article04. Do prediction-market traders act on what they say?
Connecting comments, positions, and later behavior without treating text labels as identity.
Read article05. I tried reinforcement learning too
A TradeMaster-related experiment, five seeds, and a small gain that did not survive its checks.
Read article06. The kimchi premium was more complicated than I expected
FX, funding, paired orders, and the capital consumed by a second strategy.
Read article07. Could I collect funding and wait?
Positive funding receipts did not guarantee positive returns after basis movement and costs.
Read article08. I revisited established crypto factors
Tradable shorts, changing universes, and a cost-aware ensemble that selected no winner.
Read article09. When related coins drifted apart
Residuals, hedge ratios, sizing, and a recent DRY campaign with a narrower purpose.
Read article10. Was trading an exchange announcement already too late?
Publication timestamps, delayed entries, and a feed that remained useful even without an alpha.
Read article11. I studied TRUMP and MELANIA after the fact
A retrospective question expanded into hundreds of contracts and a failed continuation test.
Read article12. I followed Korean buyback disclosures
Event timing, liquidity, and what happened when I increased position sizes.
Read article13. I looked for other opportunities in Korean equities
Screening, flow signals, SPACs, rights, and the difference between zero and unmeasured.
Read article14. What if a stock and its future have different prices?
Contract units, cash and collateral, reconciliation, and the unfinished quote-reception problem.
Read article15. Gold had different prices in different markets
Currency, grams, ounces, and redemption conditions changed the apparent gap.
Read article16. I compared crypto futures maturities and collateral
Five relative-value branches, small samples, and why similar labels did not mean identical payoffs.
Read article17. I looked for differences in option prices
Replication, volatility surfaces, and the risk between separately filled legs.
Read article18. Looking for inconsistent prices and skilled wallets
A semantic-pricing pilot lost its apparent edge to fees; wallet-copy work remained unfinished.
Read article19. What if I compared prediction markets with crypto options?
Three research lanes, fresher prices, and a large simulated return with serious capacity caveats.
Read article20. Building a research engine: what should AI handle?
Roles, deterministic calculations, memory, retries, and the moments that needed intervention.
Read article21. Building a faster backtest engine I could trust
Three clocks, shared capital, compatible caches, and checks beyond headline returns.
Read article22. Building an operating engine: after leaving the bots on
DRY and REAL, failed cancellation, late fees, and why a running process was not enough.
Read articleInvesting on my own: a 22-part research journal
Twenty-two notes on my solo investment research: ideas, failed tests, data limits, and operating experience.
Read articleHello, world
A small first note.
Read article