Mid-Career Reflections on Quant Finance

I worked 2017-2024 at Two Sigma¹, a systematic hedge fund. I loved working there, and was sad to leave.

2020-2024 my team and I founded and scaled the industry’s first (afaik) “systematic buy-side”² alpha capture business.

I’m now on non-compete leave through December 2025, blessed with time and hindsight. Here are some lessons I plan to take with me.

1. Paranoia > optimism

Default tech mindset: “How do we ship and scale this?”

Default quant mindset: “Why does this almost certainly fail?”

This is healthy paranoia. Many strategies plausibly generate value. Meanwhile, the market turns every alpha into beta into noise.

A checklist I learned the hard way:

• Nothing works unless already proven and re‑proven on trusted out‑of‑sample.
• Anything that did work probably stopped last week.
• Everything still working is arbing me.

2. Scope the entire decision space.

“Think systematically” really means “list every option”.

For example, consider whether to de-mean a feature:

• Cross-sectionally? Longitudinally? Both?
• Bucketed by another feature, or raw?
• Simple, windowed, or decayed? What kind of decay?
• Pre- or post- factor neutralization?

How many of these variations should I try? How do I pick the winner? Should I pick the winner the same way next time?

Every decision is like this, with layers and layers of choices. Dimensionalization helps.

3. Most of the time, Long > Neutral

When the markets are up (true most of the time), quants miss out. Market neutrality underperforms. Tech Equity > Quant Bonuses over the last 2 decades. And it’s more fun to root for everyone to win.

When the markets are down, everyone is worried, quants not excepted. But they are still making money.

4. Diversification → Optionality, but not vice versa.

10 orthogonal alpha strategies ≈ long 10 cheap call options³.
Buying 10 random call options does not recreate 10 orthogonal strategies.

Diversification comes from true differentiation in the source of return. Buying 10 tickets to the same lottery is not diversifying. Source independence is the scarce commodity.

5. Glory half-life ≈ 1hr.

The past is data. The present is noise. The future is revenue.

Decay starts immediately after release. Keep building; rest is expensive.

6. Failure is overdetermined and asymmetric.

Very few things unexpectedly go right: surprise usually hurts. And usually you can’t retire after one lucky call.

But any one thing that goes wrong, even (especially) unrelated to investments and ideas, can lead to ruin.

7. Agents do what rewards them.

There is not much confusion about what motivates people in quant finance. This is helpful for modeling, but brutal for trust.

Helpful, because if I know an agent acts in their own interest, their actions are a strong signal of what their interests are.

Brutal, because if I know anyone will take advantage of my weakness, I need to be on guard at all times.

Legalese is the duct tape over that bounded distrust.

8. Everything is a secret.

There will probably not be a DeepSeek moment in quant finance. No hedge fund is open-sourcing their alphas. Information is too valuable:

• Knowing what someone is doing = Knowing where they see value
• Knowing what they are not doing = Knowing where they don’t see enough value
• Knowing what failed = Knowing there probably isn’t value there

9. Reputation overcomes uncertainty.

If nothing can be trusted without a track record, and everything is a secret, how does anything get started?

Reputation is how. Reputation allows your model to have a lower discount rate, more optimistic scenario analyses, and stronger priors.

Not to mention that reputation helps you execute too.

10. The Sharpe-Capacity frontier

One of the best things about a high Sharpe strategy is that I know quickly if it stops working. Sharpe 5 means I cut after a few losing days in a row. At Sharpe 1, I might wait out years of losses.

But I can’t scale Sharpe 5 as big as Sharpe 1. Often the Sharpe‑1 behemoth out‑earns the Sharpe‑5 jewel in dollar terms. And I eat $ PNL, not σ‑adjusted virtue.

11. Opportunity Cost > Financial Cost

Every green‑lit project blocks a better unseen one and mortgages bandwidth to execute. Budget attention, not just capital.

12. Decision Quality >> Prediction Accuracy

The important thing is the buy/sell decision, not the accuracy of the input:

• I lose from a perfectly accurate KPI prediction if it causes me to buy a stock whose price is about to go down.
• I lose from a perfectly accurate price prediction if it causes me to buy a stock whose price will increase less than trading costs.

Accuracy is helpful, but the decision pays the bills⁴.

13. Friends >>> Allies

Allies are aligned: they pursue shared interests or goals. When context changes, alliances shift.

Friends stick around. But it’s hard to make friends when everyone (rationally) fears vulnerability.

This contrasts with the lesson imparted by many strategy games, where Ally is the highest relationship tier, while Friend happens early.

Coda

Your time, tools, and ideas are worth exactly what you can generate with them. If someone will pay more than that, consider it a blessing. And then price your optionality.

What heuristics keep you solvent? Leave a comment; I read everything.

1

Opinions solely mine; errors solely mine. Nothing here is investment advice.

2

“systematic buy-side” ≠ “systematic sell-side” ≠ “discretionary buy-side”

3

A good payoff profile for an alpha strategy is call-option-like: bounded downside, positive mean, fat right tail.

4

Of course, over many decisions, being more accurate should help. But accuracy only compounds if you survive.