EP05 - Quant Risk

Risk management for quantitative trading. Learn about leverage limits, diversification, systematic risk, Kelly Criterion, and how to survive in volatile markets without blowing up.

Tweet coming soon

I've done it, Kurumi. I solved the market.

I'm using **30x Leverage**. Every 1% the market moves up, I make 30%!

And if it moves down?

My model has a 99% win rate! It's an Arbitrage strategy!

I'm betting that bond yields in Italy and Germany will converge. It's mathematical destiny!

I am a genius. Just like those guys at **LTCM**.

Did you just use "LTCM" as a *positive* example?

Yeah! Nobel prize winners! They made billions!

LTCM made billions. And then they lost **$4.6 Billion in four months** and almost destroyed the global economy because they confused **Probability** with **Reality**.

Retail traders like you optimize for **Returns** (How much money I make).

Insurance companies optimize for **Risk** (How do I not lose).

Quants optimize for **Risk-Adjusted Returns**.

Speed (Return) is useless without Suspension (Risk Management).

With 30x leverage, a **3.3% drop** wipes you out.

But the market never drops 3% in a day! Not for these assets!

It hasn't... *lately*.

You are betting that the future looks exactly like the past.

That is **The Turkey Illusion**. The turkey is fed every day, so its statistical model says "The farmer loves me." Until Thanksgiving.

Okay, so I need to measure the risk. I'll use the **Sharpe Ratio**!

Return divided by Volatility. If Sharpe is high, I'm good!

Sharpe is the industry standard. But it has a flaw.

It punishes *Upside* volatility too. If you win big unexpectedly, Sharpe goes down.

Sortino only punishes *Downside* Deviation. It's better.

But ratios are just numbers. They assume the distribution is a Bell Curve.

Financial markets have **Fat Tails**.

Wait... 2% decline in SPX? Its fine. It's just a 2-sigma event!

You are levered 30x. So a 2% decline in the index is 60% of your equity.

The best part? The more you're margin called, the faster the market falls. What started as a 2-sigma event cascades into a 6-sigma event.

I'm getting margin called! I keep hitting my stop losses in each circuit breaker! I'm never going to dinner parties with Bill Hwang ever again!

I lost it all.

I made 400% return for 3 years... and lost 100% in 3 days.

The arithmetic math checks out, but the geometric math doesn't.

This is the **Absorbing Barrier**.

In a video game, if you die, you respawn.

In finance, if you hit zero, you are out of the game forever. You cannot compound zero. You are **ruined**.

I was picking up pennies in front of a steamroller...

It's a classic strategy. High probability of small gain. Tiny probability of total death.

It works great... until the steamroller speeds up.

So... how do Quants survive?

We assume the steamroller is invisible and always right behind us.

We cap our leverage. We constantly stress-test for "What if correlation goes to 1?"

If the risk gets too high, we **De-leverage**. We sell *before* the crash.

So you intentionally make *less* money than you could?

We optimize for **Survival**.

The winner isn't the one with the most money at the peak. It's the one who still has money at the trough.

Calculate the **Geometric Mean**. Not the Arithmetic Mean.

Fine. I'll reduce leverage to 2x.

Leverage does not matter. What matters is your Kelly.

Kelly Criterion? What is that?

It was explained in the previous episode.

Fine. Finance is risky. If I can't maximize returns, then I must minimize risk. Diversification! The only free lunch in finance!

If I buy 1 stock, it's risky. If I buy 5,000 stocks, my risk drops to zero! I'm going to buy every asset on Earth!

The benefit of diversification is **Logarithmic**. The cost is **Linear**.

Look at the curve. Moving from 1 stock to 20 stocks reduces your Idiosyncratic Risk by ~70%.

Moving from 20 to 1,000? You gain almost nothing.

But for every new ticker you add, you pay transaction fees. Data feed costs. Complexity risk.

At a certain point, you are just paying fees to hug the index. You become a **Closet Index Fund**.

But... why does it flatten? Why doesn't risk go to zero?

If I own everything, shouldn't I be immune to individual failures?

You can diversify away the fire in one house.

You cannot diversify away the **Earthquake**.

Interest Rates. Inflation. Nuclear War.

These affect *all* assets simultaneously.

That flat line is the **Floor**. You cannot go below it.

So... there is a minimum amount of danger I must accept?

Yes. And that floor defines your **Max Leverage**.

If the Earthquake (Systematic Risk) can drop the market by 20%...

Then I can't use more than 5x leverage.

Because $5 imes 20% = 100%$. You are dead.

I was betting that the Earthquake was impossible.

You were betting that history stopped happening.

And here is the scary part. The **Floor Moves**.

The floor moves?

If 30% of the S&P 500 is just 5 tech companies...

Then "Systematic Risk" is just "Tech Risk" in a trench coat.

That's right, your "Diversified Portfolio" is actually just a bet on Silicon Valley.

If a regulation hits Tech, the *entire market* drops. The floor falls out from under you. Your "Safe Leverage" of 5x suddenly becomes a death sentence.

This explains why math models fail. They assume the floor is solid concrete. But it's actually ice that gets thinner when things get hot.

It gets worse.

Retail thinks of stocks as bouncing balls in a box. Independent.

Engineers know they are **Dominoes**.

Domino 1 has a glitch. It goes offline for 24 hours.

Domino 2 can't process payments. They miss their monthly loan payment.

Domino 3 (The Bank) sees a spike in defaults. Their risk model panics.

The Bank stops lending to *everyone*.

Suddenly, a coffee shop in Texas can't get a loan to fix their oven.

Because a server in Virginia glitched?

Because the system is **Tightly Coupled**.

Risk models underestimate this?

Standard models (Gaussian) assume events are independent.

They don't model the **Counterparty Cascade**.

In a crisis, your "hedge" (Safety Asset) is someone else's "liability."

When they go bankrupt, your safety net vanishes.

That is why **Correlations go to 1**.

So... calculating volatility isn't enough.

We have to calculate the *breaking point* of the web.

That's why we have **Risk Capacity**.

It's not "How much *can* I leverage?"

It's "How much pain can I take when the web snaps?" Only a well capitalized bank can survive the earthquake.

I'm turning it down.

I'm diversifying up to the point of diminishing returns... and then stopping.

And I'm keeping a cash buffer for the Earthquake.

Stop trying to optimize the last 0.1% of profit, and start optimizing for structural integrity.

The real alpha is not in how smart your predictions are. It's how well you've built a financial infrastructure that serves the global maritime order.