"Entropic Value-at-Risk parity for tempered stable returns" analysis
https://arxiv.org/pdf/2609.11905 .
Based on the research paper "Entropic Value-at-Risk parity for tempered stable returns" (arXiv:2609.11905), here are the top 5 points explained in simple, easy-to-understand language:
- Building Smarter Investment Portfolios Using Advanced Risk Management:The paper focuses on portfolio management—specifically how to build investment portfolios where every asset contributes equally to the overall risk (known as Risk Parity or Equal Risk Contribution).
Instead of just looking at standard risk measures, it uses a sophisticated risk metric called Entropic Value-at-Risk (EVaR), which is much better at accounting for rare, extreme market crashes (tail risks). - Handling Real-World Financial "Surprises" (Tempered Stable Returns):Standard financial models usually assume asset returns follow a neat, bell-curve pattern (Gaussian/Normal distribution). However, real-world markets experience sudden, sharp spikes and drops. This paper uses tempered stable models, which are mathematical formulas designed to accurately capture these harsh, real-world market behaviors and heavy-tailed shocks.
- Isolating True Risk from Expected Growth:When calculating EVaR, the math can sometimes get messy by mixing up how much an asset is expected to grow (its expected return/location term) with how risky it actually is. The author successfully created a mathematical method ("EVaR-deviation") to cleanly separate baseline growth from actual risk, making the risk assessment much more accurate.
- Connecting Back to Traditional Methods:A great feature of this new model is that if markets were behaving like a simple bell-curve (Gaussian returns), this complex new EVaR method smoothly simplifies down to the classic volatility-based risk parity models that Wall Street has used for decades. This proves the new method is a natural, more powerful upgrade to older techniques.
- Proven Success in Actual Market Tests:When tested across multiple different investment universes, the portfolios built uing this new EVaR approach consistently outperformed basic, equal-weighted portfolios. Specifically, they achieved positive Sharpe ratio differences, meaning they delivered better risk-adjusted returns (more profit per unit of risk taken).
DEEPSEEK:
# EVaR Risk Parity for Tempered Stable Returns — Explained Simply
*Blog post draft — ready to copy into Google Blogger*
---
## Introduction: How Should You Split Your Money?
Imagine you have ₹100,000 to invest. You want to put it into different things — stocks, bonds, gold, real estate. The big question is:
**How much money should go into each one?**
Most people use a simple rule: "Put less money in the scary stuff." But what does "scary" mean? Most methods only look at how much prices bounce up and down. They treat a big gain and a big loss the same way.
That's a problem. Because **crashes** are what really hurt you. A 40% drop is not the same as a 40% jump.
This blog explains a research paper by **Jaehyung Choi** that builds a smarter way to split your money. It uses a risk measure called **EVaR** and realistic models called **tempered stable** returns.
Let's break it down into easy pieces.
---
## 1. What is Risk Parity?
**Risk parity** is a simple idea:
> Every asset in your portfolio should bring the **same amount of risk**.
Example:
- Asset A is very risky (like crypto)
- Asset B is very safe (like government bonds)
Risk parity says: put **less money** in A, **more money** in B, so both bring the **same risk** to your total portfolio.
There are two versions:
**IRP (Inverse Risk Parity):**
Weight goes **opposite** to risk.
Formula:
```
w_i = (1 / σ_i) / (sum of 1 / σ_j for all assets)
```
In plain words:
- If an asset is very risky → give it a small weight
- If an asset is very safe → give it a big weight
**ERC (Equal Risk Contribution):**
Adjust weights so every asset contributes **exactly the same amount of risk**.
Both are popular. But they usually use **volatility** (bounciness) as the measure of risk.
**Problem:** Volatility treats good days and bad days the same. It doesn't focus on crashes.
---
## 2. What is EVaR?
**VaR (Value-at-Risk)** answers: "How much could I lose on a bad day?"
Example: "With 95% confidence, I won't lose more than ₹10,000 tomorrow."
**Problem with VaR:** It doesn't tell you **how bad** the bad day is. You could lose ₹10,001 or ₹10,00,000 — VaR doesn't say.
**EVaR (Entropic Value-at-Risk)** is a smarter version. It looks at the **average of the worst cases**, not just the cutoff line.
Formula:
```
EVaR = smallest value of [ log M(u) - ln(η) ] / u
over all u > 0
```
Where:
- `M(u)` = moment-generating function (a math tool that describes the whole distribution)
- `η` = small number like 0.05 (the "bad tail" probability)
- The answer = a number in dollars or percent
**Why EVaR?**
- It is **coherent** (mathematically well-behaved)
- It **cares about the tail** (the crashes)
- It works well with optimization
---
## 3. What are "Tempered Stable" Returns?
Real stock returns are **not** a perfect bell curve.
They have:
- **Fat tails** — crashes happen more often than a bell curve predicts
- **Skew** — bad days can be worse than good days are good
**Tempered stable** models are a family of formulas that fit real returns much better.
The paper uses two types:
- **MNTS** = Multivariate Normal Tempered Stable
- **ICA** = Independent Component Analysis (break returns into independent pieces, each tempered stable)
You don't need the deep math. Just know: **these are more realistic than assuming a bell curve.**
---
## 4. The Key New Idea: EVaR Contribution
To do ERC, you must know: **how much does each asset contribute to total portfolio EVaR?**
The paper derives this. It's called the **Euler contribution**:
Formula:
```
EVaRC_i = w_i × (change in total EVaR / change in w_i)
```
In plain words:
> **Contribution of asset i = its weight × how much total risk changes when you nudge that weight.**
Think of it like this: "If I add 1% more of asset i, how much more risk does the whole portfolio get?"
**ERC condition:** make all these contributions equal.
```
EVaRC_i = EVaRC_j for all assets i and j
```
In plain words: **every asset contributes the same risk.**
---
## 5. The "Deviation" Trick — Separating Average from Risk
EVaR includes the **average return** (the "location"). But that's not really risk — that's reward.
So the paper defines **EVaR deviation**:
```
EVaR deviation = EVaR of (returns with the mean subtracted)
```
In plain words: **subtract the average first, then compute EVaR.** Now you're looking at pure risk.
**The clean split:**
```
Total EVaR = (negative average return) + (pure risk)
```
Same for contributions:
```
Total contribution = (average part) + (pure risk part)
```
This is elegant. It separates **reward** from **risk**.
---
## 6. Beautiful Result: The Gaussian Case
If returns are **Gaussian** (bell curve), then:
```
EVaR deviation = constant × volatility
```
So EVaR-deviation IRP and ERC give **exactly the same weights** as old-fashioned volatility IRP and ERC.
**Translation:** the new method is a **generalization** — it becomes the old method when returns are normal.
This is important. It means the new method is not a random alternative. It is a **natural extension** of what people already do.
---
## 7. How They Tested It
Three groups of assets:
**XASSET:** 7 ETFs
- US stocks (VTI)
- Foreign stocks (EFA)
- Emerging markets (VWO)
- Real estate (VNQ)
- Bonds (AGG)
- Inflation bonds (TIP)
- Gold (GLD)
**MOM10:** 10 momentum-sorted stock portfolios
**SECTOR:** 9–11 US sector ETFs
- Tech, health, energy, finance, etc.
**Setup:**
- Monthly rebalancing
- 12 months of past data to fit the model
- 95% confidence (η = 0.05)
- Long-only, fully invested
**Compared against:**
- Equal weight (1/N)
- Gaussian EVaR benchmark
- Matched CVaR portfolios
---
## 8. What They Found
**vs Equal Weight:**
- ERC beat equal weight in **all 9** comparisons (Sharpe ratio)
- IRP beat equal weight in **8 out of 9**
- Best result: XASSET MNTS ERC → **+0.246 Sharpe** (p = 0.011)
**vs CVaR:**
- Under direct MNTS: EVaR ≈ CVaR (basically the same)
- Under ICA-based ERC: differences appear, but **signs flip** across universes
- Sometimes EVaR wins, sometimes CVaR wins
- **No clear winner**
**Transaction costs:**
- ICA-based EVaR-ERC trades **a lot more** (higher turnover)
- This eats into the gains
- At 25 basis points cost, most advantages shrink and become insignificant
**Deviation vs raw:**
- Removing the mean term had **almost no effect** on Sharpe ratios
- So the location term wasn't the main driver
---
## 9. The Honest Bottom Line
**What they achieved:**
- A rigorous new framework for EVaR-based risk parity
- Derived all the needed formulas for tempered stable models
- Showed ERC beats equal weight in their tests
**What they admit:**
- EVaR is **not always better** than CVaR
- Results depend on the universe, the period, and the model
- Turnover and transaction costs matter a lot
- This is a **methodological contribution**, not a guaranteed money-maker
**One-sentence summary:**
> The paper builds a smarter risk parity method using EVaR and realistic fat-tailed models, shows it beats equal weight, but honestly reports that it doesn't consistently beat the simpler CVaR approach.
---
## Key Formulas — All in One Place
| Concept | Formula | Plain English |
|---|---|---|
| IRP weight | w_i = (1/σ_i) / Σ(1/σ_j) | Weight goes opposite to risk |
| EVaR | min over u of [log M(u) − ln η] / u | Smallest value of a special ratio |
| EVaR contribution | EVaRC_i = w_i × ∂D/∂w_i | Weight × marginal risk |
| ERC condition | EVaRC_i = EVaRC_j | All contributions equal |
| EVaR split | D = −wᵀμ + D^c | Total = location + pure risk |
| Gaussian case | D^c = k_η × √(wᵀΩw) | EVaR deviation = constant × volatility |
| ERC share | EVaRC_i = (1/N) × D | Each asset gets 1/N of total risk |
---
## The Whole Paper in One Picture
```
You have money to invest.
↓
Old way: look at volatility (bounciness).
↓
New way: look at EVaR (crash risk) + tempered stable (realistic returns).
↓
Split money so every asset brings equal crash risk (ERC).
↓
Result: beats equal weight. But not always better than CVaR.
```
---
## Final Thoughts
This paper is a **methodological contribution**. It doesn't promise you will get rich. It doesn't say EVaR is always better than CVaR. It says:
> "Here is a rigorous way to do risk parity with EVaR and realistic models. It works well in our tests. But be careful — results depend on your universe, your period, and your costs."
That honesty is what makes it good research.
If you are building a portfolio and want to go beyond simple volatility-based risk parity, this paper gives you the tools. But always remember: **backtest on your own data, include transaction costs, and don't trust any single method blindly.**
---
*Disclaimer: This blog is for educational purposes only. It is not financial advice. Always do your own research before investing.*
---
**Reference:**
Choi, J. (2026). *Entropic Value-at-Risk parity for tempered stable returns.* arXiv:2609.11905.
