Download docx - Fight Finance

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Discrete choice wikipedia , lookup

Choice modelling wikipedia , lookup

Rate of return wikipedia , lookup

Modified Dietz method wikipedia , lookup

Transcript
Prices are Log-Normally Distributed
Since security prices (p) cannot fall below zero due to limited
liability, they're normally assumed to be log-normally
distributed with a left-most value of zero.
๐‘๐‘ก ~ ๐‘™๐‘›๐‘(๐‘š๐‘’๐‘Ž๐‘›, ๐‘ฃ๐‘Ž๐‘Ÿ๐‘–๐‘Ž๐‘›๐‘๐‘’)
0 โ‰ค ๐‘๐‘ก < โˆž
Of course, in reality this assumption of log-normal prices (and
normal continuously compounded returns) is broken, but
keeping it makes the mathematics tractable.
Revised KW: 14.5.17
2
GDR's and NDR's are also Log-Normally
distributed
Since gross discrete returns (GDR's) and net discrete returns
(NDR's, also known as effective returns) are linear functions of
the price, they are also log-normally distributed.
GDR's have a left-most point of zero similarly to prices.
๐บ๐ท๐‘… ~ ๐‘™๐‘›๐‘(๐‘š๐‘’๐‘Ž๐‘›, ๐‘ฃ๐‘Ž๐‘Ÿ๐‘–๐‘Ž๐‘›๐‘๐‘’)
0 โ‰ค ๐บ๐ท๐‘…๐‘กโ†’๐‘ก+1 < โˆž
NDR have a left-most point of negative one.
๐‘๐ท๐‘… ~ ๐‘™๐‘›๐‘(๐‘š๐‘’๐‘Ž๐‘›, ๐‘ฃ๐‘Ž๐‘Ÿ๐‘–๐‘Ž๐‘›๐‘๐‘’)
โˆ’1 โ‰ค ๐‘๐ท๐‘…๐‘กโ†’๐‘ก+1 < โˆž
3
LGDR's are Normally distributed
๐‘๐‘ก
Log gross discrete returns (๐ฟ๐บ๐ท๐‘… = ln ( )) are normally
๐‘0
distributed and they are unbounded in the positive and
negative directions.
๐ฟ๐บ๐ท๐‘… ~ ๐‘(๐‘š๐‘’๐‘Ž๐‘›, ๐‘ฃ๐‘Ž๐‘Ÿ๐‘–๐‘Ž๐‘›๐‘๐‘’)
โˆ’โˆž โ‰ค ๐ฟ๐บ๐ท๐‘…๐‘กโ†’๐‘ก+1 < โˆž
4
5
AAGDR = Mean GDR
Arithmetic average of the gross discrete returns (AAGDR)
๐‘‡
๐ด๐ด๐บ๐ท๐‘… 0โ†’๐‘‡
1
๐‘๐‘ก
= .โˆ‘(
)
๐‘‡
๐‘๐‘กโˆ’1
๐‘ก=1
๐‘1 ๐‘2
๐‘
+ + โ‹ฏ+ ๐‘‡
๐‘0 ๐‘1
๐‘๐‘‡โˆ’1
=
๐‘‡
๐บ๐ท๐‘…0โ†’1 + ๐บ๐ท๐‘…1โ†’2 + โ‹ฏ + ๐บ๐ท๐‘…๐‘‡โˆ’1โ†’๐‘‡
=
๐‘‡
6
GAGDR = Median GDR
Geometric average of the gross discrete returns (GAGDR):
1
๐‘‡
๐‘‡
๐บ๐ด๐บ๐ท๐‘…0โ†’๐‘‡
๐‘๐‘ก
= (โˆ (
))
๐‘๐‘กโˆ’1
๐‘ก=1
= (๐บ๐ท๐‘…0โ†’1 × ๐บ๐ท๐‘…1โ†’2 × โ€ฆ ×
1
๐บ๐ท๐‘…๐‘‡โˆ’1โ†’๐‘‡ )๐‘‡
1
๐‘‡
๐‘1 ๐‘2
๐‘๐‘‡
= ( × × โ€ฆ×
)
๐‘0 ๐‘1
๐‘๐‘‡โˆ’1
1
๐‘๐‘‡ ๐‘‡
=(
๐‘0
) =
1
(๐บ๐ท๐‘…0โ†’๐‘‡ ) ๐‘‡
7
LGAGDR = AALGDR = Mean, Median and
Mode LGDR
The log of the geometric average of the gross discrete returns
(LGAGDR) is:
๐‘‡
๐ฟ๐บ๐ด๐บ๐ท๐‘…0โ†’๐‘‡
1
๐‘‡
๐‘๐‘ก
= ln ((โˆ (
)) )
๐‘๐‘กโˆ’1
๐‘ก=1
1
๐‘๐‘‡
1
๐‘‡
๐‘0
๐‘‡
= . ln ( ) = . ln(๐บ๐ท๐‘…0โ†’๐‘‡ )
8
Due to the properties of the logarithmic function, the log of the
geometric average of the gross discrete returns (LGAGDR)
equals the arithmetic average of the log gross discrete returns
(AALGDR):
๐‘‡
๐ฟ๐บ๐ด๐บ๐ท๐‘…0โ†’๐‘‡
1
๐‘‡
๐‘๐‘ก
= ln ((โˆ (
)) )
๐‘๐‘กโˆ’1
๐‘ก=1
1
๐‘1 ๐‘2
๐‘๐‘‡
= . ln ( × × โ€ฆ ×
)
๐‘‡
๐‘0 ๐‘1
๐‘๐‘‡โˆ’1
1
๐‘1
๐‘2
๐‘๐‘‡
= . (ln ( ) + ln ( ) + โ‹ฏ + ln (
))
๐‘‡
๐‘0
๐‘1
๐‘๐‘‡โˆ’1
9
๐‘‡
1
๐‘๐‘ก
= . โˆ‘ (ln (
))
๐‘‡
๐‘๐‘กโˆ’1
๐‘ก=1
= ๐ด๐ด๐ฟ๐บ๐ท๐‘…0โ†’๐‘‡
Since the logarithms of the gross discrete returns (LGDRโ€™s) are
normally distributed, the LGAGDR and AALGDR are equal to
the mean, median and mode of the LGDRโ€™s.
10
SDLGDR
The arithmetic standard deviation of the log gross discrete
returns (SDLGDR) is defined as:
๐‘‡
2
1
๐‘๐‘ก
๐‘†๐ท๐ฟ๐บ๐ท๐‘… = ๐œŽ = โˆš . โˆ‘ ((ln (
) โˆ’ ๐ด๐ด๐ฟ๐บ๐ท๐‘…0โ†’๐‘‡ ) )
๐‘‡
๐‘๐‘กโˆ’1
๐‘ก=1
Since ๐ด๐ด๐ฟ๐บ๐ท๐‘…0โ†’๐‘‡ = ๐ฟ๐บ๐ด๐บ๐ท๐‘…0โ†’๐‘‡ = ln(๐บ๐ด๐บ๐ท๐‘…0โ†’๐‘‡ ), then:
๐‘‡
1
๐บ๐ท๐‘…๐‘กโˆ’1โ†’๐‘ก 2
๐‘†๐ท๐ฟ๐บ๐ท๐‘… = โˆš . โˆ‘ ((ln (
)) )
๐‘‡
๐บ๐ด๐บ๐ท๐‘…0โ†’๐‘‡
๐‘ก=1
11
The geometric standard deviation of the gross discrete
returns is defined as the exponential of the arithmetic
standard deviation of the log gross discrete returns (SDLGDR).
GeometricSDGDR = exp(SDLGDR)
=
2
๐บ๐ท๐‘…๐‘กโˆ’1โ†’๐‘ก
๐‘‡
exp (โˆš . โˆ‘๐‘ก=1 ((๐‘™๐‘› (
)) ))
๐‘‡
๐บ๐ด๐บ๐ท๐‘…
1
0โ†’๐‘‡
12
AAGDR = exp(AALGDR + SDLGDR2/2)
There is an interesting relationship between the mean or
arithmetic average gross discrete return (AAGDR) and the
arithmetic average of the log gross discrete return (AALGDR or
continuously compounded return).
๐ด๐ด๐บ๐ท๐‘… =
๐‘†๐ท๐ฟ๐บ๐ท๐‘… 2
๐‘’ ๐ด๐ด๐ฟ๐บ๐ท๐‘…+ 2
๐‘†๐ท๐ฟ๐บ๐ท๐‘…2
= exp (๐ฟ๐บ๐ด๐บ๐ท๐‘… +
)
2
๐‘†๐ท๐ฟ๐บ๐ท๐‘…2
= exp(๐ฟ๐บ๐ด๐บ๐ท๐‘…) × exp (
)
2
13
๐‘†๐ท๐ฟ๐บ๐ท๐‘…2
๐ด๐ด๐บ๐ท๐‘… = ๐บ๐ด๐บ๐ท๐‘… × exp (
)
2
Where the variables above are defined as:
๐‘‡
2
1
๐‘๐‘ก
๐‘†๐ท๐ฟ๐บ๐ท๐‘… = โˆš . โˆ‘ ((ln (
) โˆ’ ๐ด๐ด๐ฟ๐บ๐ท๐‘…) )
๐‘‡
๐‘๐‘กโˆ’1
๐‘ก=1
๐‘‡
1
๐‘๐‘ก
๐ฟ๐ด๐ด๐บ๐ท๐‘… = ln ( . โˆ‘ (
))
๐‘‡
๐‘๐‘กโˆ’1
๐‘ก=1
๐‘‡
1
๐‘๐‘ก
๐ด๐ด๐ฟ๐บ๐ท๐‘… = . โˆ‘ (ln (
)) = ๐ฟ๐บ๐ด๐บ๐ท๐‘…
๐‘‡
๐‘๐‘กโˆ’1
๐‘ก=1
14
Necessary assumptions:
๏‚ท LGDR's (also called continuously compounded returns)
must be normally distributed and therefore the GDRโ€™s are
log-normally distributed.
๏‚ท If the returns (LGDR's) are of a portfolio, this equation only
holds if the weights are continuously rebalanced.
15
Mean GDR โ‰ฅ Median GDR
๐‘†๐ท๐ฟ๐บ๐ท๐‘…2
๐ด๐ด๐บ๐ท๐‘… = ๐บ๐ด๐บ๐ท๐‘… × exp (
)
2
This is equivalent to:
SDLGDR2
Mean GDR = Median GDR × exp (
)
2
So the Mean GDR (or AAGDR) is always greater than or equal
to the Median GDR (or GAGDR) because exp (
SDLGDR2
2
) is
always greater than or equal to one.
๐Œ๐ž๐š๐ง ๐†๐ƒ๐‘ โ‰ฅ ๐Œ๐ž๐๐ข๐š๐ง ๐†๐ƒ๐‘
or
๐€๐€๐†๐ƒ๐‘ โ‰ฅ ๐†๐€๐†๐ƒ๐‘
This can be shown intuitively using a graph.
16
Image author:
Olivier de La
Grandville, from
his book โ€˜Bond
Pricing and
Portfolio
Analysisโ€™, 2001.
17
18
Why is this interesting?
Future prices are thought to be log-normally distributed.
The mean and median future prices will be different!
SDLGDR2
(๐ด๐ด๐ฟ๐บ๐ท๐‘…+
).๐‘ก
2
.๐‘’
๐‘€๐‘’๐‘Ž๐‘›๐‘ƒ๐‘ก = ๐‘ƒ0
= ๐‘ƒ0 . (1 + ๐ด๐ด๐‘ต๐ท๐‘…)๐‘ก
๐‘€๐‘’๐‘‘๐‘–๐‘Ž๐‘›๐‘ƒ๐‘ก = ๐‘ƒ0 . ๐‘’ ๐ด๐ด๐ฟ๐บ๐ท๐‘….๐‘ก =
= ๐‘ƒ0 . ((1 +
= ๐‘ƒ0 . ๐ด๐ด๐‘ฎ๐ท๐‘…๐‘ก =
๐‘†๐ท๐ฟ๐บ๐ท๐‘… 2
๐‘ƒ0 . ๐ด๐ด๐‘ฎ๐ท๐‘…๐‘ก . ๐‘’ โˆ’ 2 .๐‘ก
๐‘†๐ท๐ฟ๐บ๐ท๐‘… 2
๐ด๐ด๐‘ต๐ท๐‘…). ๐‘’ โˆ’ 2 )
๐‘ก
This has important implications.
19
Mean versus Median of a Log-normally
distributed variable
The mean of a log-normally distributed variable is always
higher than the median.
Price, gross discrete returns and net discrete returns (effective
returns) are log-normally distributed.
Due to the right tail in the log-normal distribution, there are
some very high values which increase the mean but have less
effect on the median.
The chance of achieving a GDR, NDR or price above the
mean is less than 50%.
20
The chance of achieving a return above the median is exactly
50%. The median is the 50th percentile.
21
Time Weighted Average = GAGDR = Median
GDR
The โ€˜time-weighted averageโ€™ GDR or NDR is synonymous with
the geometric average GDR or NDR.
Over time, the chance of achieving or exceeding the median
price, GDR or NDR will always be 50%.
22
Ensemble Average = AAGDR = Mean GDR
Over time, the chance of achieving or exceeding the mean
price, GDR or NDR will get smaller and smaller. As time
approaches infinity, the chance will approach zero.
This makes sense because there are a small number of shares
that will do exceptionally well. Think of Berkshire Hathaway,
Google or Blackmores which have had incredibly high returns.
There is no limit to how high prices, GDRโ€™s or NDRโ€™s can go.
However, the lowest that prices and GDRโ€™s can go is zero and
NDRโ€™s canโ€™t get lower than minus 1 which is losing 100%.
23
The arithmetic average is pulled up by the small number of
very high returns, while the median or geometric average, the
โ€˜middle valueโ€™, is not affected as much.
24