Beruflich Dokumente
Kultur Dokumente
sA
1 n
2
rcont,i mA
n i 1
Standard deviation is the square root of the mean square deviation from the arithmetic
mean mA calculated as
mA
1 n
rcont,i
n i 1
The deviation measured in volatility is the arithmetic difference of continuous returns from
the mean, the arithmetic mean.
1/7
Remember that continuous returns are called such because they assume continuous
compounding
rcont,i ln 1 rdisc,i
If we analyze continuous and discrete returns in a multi-period setting, we see that discrete
compounding is a multiplicative i.e. geometric process and continuous compounding an
additive i.e. arithmetic process
mG 1 rdisc,i
i 1
2/7
1n
n
1
ln 1 mG ln exp rcont,i
ln exp rcont,1 ln exp rcont, 2 ... ln exp rcont,n
i 1
n
ln 1 mG
1 n
rcont,i mA
n i 1
respectively
mG exp mA 1
...and
mA ln 1 mG
Therefore, the conversion formulas introduced above are not only applicable for single
period returns, but also average returns. Substituting the definition of discrete returns into
the volatility formula
1 n
2
sA
ln 1 rdisc,i ln 1 mG
n i 1
and applying the exponential function to both sides of the above expression and
subtracting one
2
1 n
1 n 1 rdisc,i
2
1 sG
exp s A 1 exp
ln 1 rdisc,i ln 1 mG 1 exp
ln
i 1
i 1
n
n
1
results in what is known as geometric volatility sG. From the above expressions, we see
that geometric volatility can be calculated directly as
2
1 n 1 rdisc,i
sG exp
ln
1
n i 1 1 mG
sG exp s A 1
s A ln 1 sG
We see that the conversion formulas not only apply to return and average returns, but also
standard deviation. In fact, it can be shown that the conversion formulas apply to any
moments1.
The central moments, to be more precise. The conversion formulas cannot be applied to the sample skewness and
excess kurtosis directly.
3/7
Note that the geometric standard deviation is not equal to the arithmetic standard deviation
calculated from discrete returns
sG
1 n
2
rdisc,i mA
n i 1
with
mA
1 n
rdisc,i
n i 1
Applying an arithmetic operation to geometric data leads to results which do not have a
meaningful financial interpretation. This is most obvious when trying to interpret the
arithmetic mean for discrete returns. Unfortunately, this calculation has been implemented
in many spreadsheets and systems.
lbA mA k s A
ubA mA k s A
For discrete returns, geometric versions of the above formulas have to be used
lbG
1 mG
1
1 sG k
ubG 1 mG 1 sG 1
k
1 10.5171%
32.9680%
1 28.4025%2
and
lbA ln 1 lbG
ubA ln 1 ubG
40% ln 1 32.9680%
60% ln 1 82.2119%
2011, Andreas Steiner Consulting GmbH. All rights reserved.
4/7
How relevant is the difference between arithmetic and geometric volatility? The smaller the
period of time over which returns are measured, the smaller the difference between
continuous and discrete returns. Therefore, the higher the data frequency, the smaller the
difference between geometric and arithmetic volatility. For example, when calculating daily
equity volatility
S&P 500 Arithmetic and Geometric Volatility
Daily Volatilities
3.50%
Arithmetic Volatility
Geometric Volatility
3.00%
2.50%
2.00%
1.50%
1.00%
0.50%
18.01.11
18.01.08
18.01.05
18.01.02
18.01.99
18.01.96
18.01.93
18.01.90
18.01.87
18.01.84
18.01.81
18.01.78
18.01.75
18.01.72
18.01.69
18.01.66
18.01.63
18.01.60
18.01.57
18.01.54
18.01.51
0.00%
The line for arithmetic and geometric volatility overlap, i.e. no relevant difference exists. If
we calculate 250 day volatilities from the same data, we get relevant differences
S&P 500 Arithmetic and Geometric Volatility
Annualized Rolling 250-Day Volatilities
70.00%
Arithmetic Volatility
Geometric Volatility
60.00%
50.00%
40.00%
30.00%
20.00%
10.00%
18.01.11
18.01.08
18.01.05
18.01.02
18.01.99
18.01.96
18.01.93
18.01.90
18.01.87
18.01.84
18.01.81
18.01.78
18.01.75
18.01.72
18.01.69
18.01.66
18.01.63
18.01.60
18.01.57
18.01.54
18.01.51
0.00%
5/7
We see that the difference between geometric and arithmetic volatility only becomes
relevant if the figures are large. This can be seen when plotting the conversion formulas
200.00%
180.00%
160.00%
140.00%
120.00%
100.00%
80.00%
60.00%
40.00%
20.00%
0.00%
0.00%
It follows that, the riskier the investment strategy analyzed (e.g. equity versus money
market), the more relevant the difference between geometric and arithmetic volatility.
Both the exponential and logarithmic functions are positive monotone. It can be shown that
this mathematical property implies that rankings are not affected whether they are based on
geometric or arithmetic volatilities. For example, if all UCITS IV funds report their SRRI
expressed in terms of arithmetic volatility, it would not change the relative riskiness of a
UCITS fund if the regulator would switch to geometric volatilities.
While geometric and arithmetic moments can be converted directly, this is not possible with
Sharpe ratios
SharpeA
m A exp m A 1 mG
SharpeG
sA
exp s A 1 sG
Interestingly, Sharpe ratio rankings are also not affected if returns are measured
geometrically and volatilities arithmetically (or vice versa). Of course, rankings are affected
when geometric Sharpe ratios is compared to arithmetic Sharpe ratios.
The arithmetic mean is always larger than the geometric mean. The same applies to
arithmetic standard deviation; it is always larger than its geometric counterpart. Individuals
typically like return and dislike volatility risk, which would explain the strong interest in
reporting geometric averages of discrete returns and absence of interest in reporting proper
geometric volatility figures.
6/7
Summary
Arithmetic and geometric averages are part of the established body of knowledge of the
investment industry. We have shown that there also are arithmetic and geometric
volatilities. The choice affects reported figures when they are large, and they never affect
the conclusions derived from comparing figures over time or portfolios: geometric and
arithmetic calculations do not affect the ranking of returns, volatilities or Sharpe ratios.
Comparisons may be distorted when mixing geometric and arithmetic calculations or
applying them in the wrong context. Unfortunately, the current industry practice of reporting
geometric average returns and arithmetic volatilities has the potential to distort risk and
return analysis. Discrete returns should be described with geometric statistics, continuous
returns with arithmetic statistics. The choice of the calculation methodology has to be
aligned with the compounding model chosen. The compounding model itself neither affects
investor wealth nor the performance of investment managers; it is merely a reporting
convention.
7/7