miryoricalc

CAGR vs Simple Average: Why Arithmetic Mean Overstates Compound Growth

When evaluating the multi-year performance of an investment portfolio or a economic metric, a common error is averaging annual return percentages with a simple arithmetic mean. If an asset rises $+50\%$ in Year 1 and falls $-50\%$ in Year 2, the arithmetic mean is $0\%$, yet $1,000$ USD initial capital becomes $1,500$ USD and then drops to $750$ USD — a real loss of $-25\%$.

The Definition of CAGR

The Compound Annual Growth Rate (CAGR) is the constant geometric rate $g$ required for a starting value $V_i > 0$ to reach a final value $V_f > 0$ over $n > 0$ periods:

g = (Vf / Vi) ** (1 / n) - 1

Mathematically, $1 + g$ is the geometric mean of the annual growth factors $(1 + r_1), (1 + r_2), \dots, (1 + r_n)$. By the AM-GM inequality, the geometric mean of non-equal positive values is strictly less than their arithmetic mean. The difference between arithmetic average and CAGR is often called the volatility drag, approximately equal to $\frac{1}{2}\sigma^2$ for small return variances.

Worked Example Over 4 Years

Consider an asset growing from $10,000$ USD to $14,641$ USD over $n = 4$ years:

Year End Value Annual Return
0 $10,000.00 —
1 $13,000.00 +30.00%
2 $11,700.00 -10.00%
3 $14,625.00 +25.00%
4 $14,641.00 +0.11%

The arithmetic average of returns is $(30 - 10 + 25 + 0.11) / 4 \approx 11.28\%$. But the true CAGR is:

(14641 / 10000) ** (1 / 4) - 1 = 10.00%

Code Edge Cases to Guard Against

  1. Zero Initial Value ($V_i = 0$): Division by zero. Compound growth cannot be computed relative to zero.
  2. Negative Final Value ($V_f < 0$): Taking a fractional power $1/n$ of a negative ratio yields NaN or complex numbers.
  3. Partial-Year Periods ($n$ in days or months): Normalize $n = \text{days} / 365.25$ rather than assuming integer years.
import math

def calculate_cagr(v_initial: float, v_final: float, years: float) -> float:
    if years <= 0:
        raise ValueError("Years must be positive.")
    if v_initial <= 0 or v_final < 0:
        raise ValueError("Initial value must be > 0 and final value >= 0.")
    if v_final == 0:
        return -1.0
    return math.expm1(math.log(v_final / v_initial) / years)

Using math.expm1(math.log(v_final / v_initial) / years) preserves floating-point precision when growth rate $g$ is close to zero.

Verification and Simulation

To test custom compounding horizons and view period-by-period growth tables, check the CAGR calculator at Miryori.


Disclosure: The author leads content at Miryori, the free online calculator suite linked above. Educational math only, not financial advice.