Characteristics of Asset Returns

The contents of this presentation are based on Tsay (2010)

Dr. Vinodh Madhavan

Associate Professor
Amrut Mody School of Management
Ahmedabad University
Contact Details: [email protected]
 Holding Period Returns – Simple Returns

 One Period Simple Return

𝑃𝑡 − 𝑃𝑡−1
𝑅𝑡 =
𝑃𝑡−1

𝑃𝑡
𝑅𝑡 = −1
𝑃𝑡−1

𝑃𝑡
1 + 𝑅𝑡 =
𝑃𝑡−1
 Holding Period Returns – Simple Returns
 Muti-period Holding Period Returns

𝑃𝑡
1 + 𝑅𝑡 𝑘 =
𝑃𝑡−𝑘

𝑃𝑡 𝑃𝑡−1 𝑃𝑡−2 𝑃𝑡−𝑘+1

1 + 𝑅𝑡 𝑘 = × × × ⋯…×
𝑃𝑡−1 𝑃𝑡−2 𝑃𝑡−3 𝑃𝑡−𝑘

1 + 𝑅𝑡 𝑘 = 1 + 𝑅𝑡 × 1 + 𝑅𝑡−1 × 1 + 𝑅𝑡−2 × ⋯ × 1 + 𝑅 𝑡−𝑘+1

𝑘−1

1 + 𝑅𝑡 𝑘 = ෑ(1 + 𝑅𝑡−𝑗 )
𝑗=0
 Holding Period Returns – Simple Returns
𝑘−1

1 + 𝑅𝑡 𝑘 = ෑ(1 + 𝑅𝑡−𝑗 )
𝑗=0

 Annualized average returns is derived as follows

1
𝑘−1 𝑘

𝐴𝑛𝑛𝑢𝑎𝑙𝑖𝑧𝑒𝑑 𝐴𝑣𝑒𝑟𝑎𝑔𝑒 = ෑ(1 + 𝑅𝑡−𝑗 ) −1

𝑗=0
 Continuously Compounded Returns

 Natural logarithm of Simple Gross Return Factor (1 + 𝑅𝑡 ) is

termed as Continuously Compounded Returns or Logarithmic
returns.

𝑃𝑡
𝑟𝑡 = ln 1 + 𝑅𝑡 = ln
𝑃𝑡−1

𝑟𝑡 = ln 𝑃𝑡 − ln 𝑃𝑡−1

𝑟𝑡 = 𝑝𝑡 − 𝑝𝑡−1
 Continuously Compounded Returns

𝑟𝑡 𝑘 = ln 1 + 𝑅𝑡 𝑘

𝑟𝑡 𝑘 = ln[ 1 + 𝑅𝑡 × 1 + 𝑅𝑡−1 × 1 + 𝑅𝑡−2 × ⋯ × 1 + 𝑅𝑡−𝑘+1 ]

𝑟𝑡 𝑘 = ln 1 + 𝑅𝑡 + ln 1 + 𝑅𝑡−1 + ln 1 + 𝑅𝑡−2 + ⋯ + ln(1 + 𝑅𝑡−𝑘+1 )

𝑟𝑡 𝑘 = 𝑟𝑡 + 𝑟𝑡−1 + 𝑟𝑡−2 + ⋯ + 𝑟𝑡−𝑘+1

 Appropriately Adjust for Interim Cash Flows
 The derivations thus far for simple and compound returns is premised on
there being no interim cash flows during the holding period.

 Should an asset offer interim cash flows – like a stock paying dividends at
the end of every year during the investor’s holding period or a bond that
pays coupons at the end of every year – then the simple and continuously
compound returns frameworks need to be adjusted appropriately.

𝑃𝑡 − 𝑃𝑡−1 + 𝐷𝑡
𝑅𝑡 =
𝑃𝑡−1

𝑃𝑡 + 𝐷𝑡
1 + 𝑅𝑡 =
𝑃𝑡−1

ln 1 + 𝑅𝑡 = ln 𝑃𝑡 + 𝐷𝑡 − ln(𝑃𝑡−1 )
 Moments of a continuous variable
𝑇
1
𝑀𝑒𝑎𝑛 ∶ 𝜇𝑥 = ෍ 𝑥𝑡
𝑇
𝑡=1

𝑇
1
𝑉𝑎𝑟𝑖𝑎𝑛𝑐𝑒: 𝜎𝑥2 = ෍ 𝑥𝑡 − 𝜇𝑥 2
𝑇−1
𝑡=1

𝑇 3
1 𝑥𝑡 − 𝜇𝑥
𝑆𝑎𝑚𝑝𝑙𝑒 𝑆𝑘𝑒𝑤𝑛𝑒𝑠𝑠: 𝑆𝑥 = ෍
𝑇−1 𝜎𝑥3
𝑡=1

𝑇 4
1 𝑥𝑡 − 𝜇𝑥
𝑆𝑎𝑚𝑝𝑒 𝐾𝑢𝑟𝑡𝑜𝑠𝑖𝑠: 𝐾𝑥 = ෍
𝑇−1 𝜎𝑥4
𝑡=1
 Moments of a continuous variable
𝑇
1
𝑀𝑒𝑎𝑛 = 𝐸 𝑥𝑡 = 𝜇𝑥 = ෍ 𝑥𝑡
𝑇
𝑡=1

𝑇
1
𝑉𝑎𝑟𝑖𝑎𝑛𝑐𝑒: 𝜎𝑥2 = 𝐸[ 𝑥𝑡 − 𝜇𝑥 2] = ෍ 𝑥𝑡 − 𝜇𝑥 2
𝑇−1
𝑡=1

3 𝑇 3
𝑥𝑡 − 𝜇𝑥 1 𝑥𝑡 − 𝜇𝑥
𝑆𝑎𝑚𝑝𝑙𝑒 𝑆𝑘𝑒𝑤𝑛𝑒𝑠𝑠: 𝑆𝑥 = 𝐸 = ෍
𝜎𝑥3 𝑇−1 𝜎𝑥3
𝑡=1

4 𝑇 4
𝑥𝑡 − 𝜇𝑥 1 𝑥𝑡 − 𝜇𝑥
𝑆𝑎𝑚𝑝𝑒 𝐾𝑢𝑟𝑡𝑜𝑠𝑖𝑠: 𝐾𝑥 = 𝐸 = ෍
𝜎𝑥4 𝑇−1 𝜎𝑥4
𝑡=1
 Normal Distribution
 Classical Finance is based on the assumption that logarithmic returns
[ln 1 + 𝑅𝑡 𝑜𝑟 𝑟𝑡 ] is normally distributed in nature.

 If logarithmic returns are normally distributed in nature, then asset

prices are termed as lognormally distributed.

 Under normal distribution assumption, both skewness 𝑆𝑥 and Excess

Kurtosis [𝐾𝑥 − 3] possess the following properties.

6
𝑆𝑥 : 𝑚𝑒𝑎𝑛 = 0 𝑎𝑛𝑑 𝑉𝑎𝑟𝑖𝑎𝑛𝑐𝑒 =
𝑇

24
𝐾𝑥 − 3: 𝑚𝑒𝑎𝑛 = 0 𝑎𝑛𝑑 𝑉𝑎𝑟𝑖𝑎𝑛𝑐𝑒 =
𝑇
 Testing for 𝑺𝒙 and [𝑲𝒙 − 𝟑]

 t-statistics for sample skewness is as follows

𝑆𝑥
𝑡=
6
𝑇

𝐻0 : 𝑆𝑥 = 0

𝐻𝑎 : 𝑆𝑥 ≠ 0

Null hypothesis is rejected if p-value is less than significance level

(let’s say 0.05)
 Testing for 𝑺𝒙 and [𝑲𝒙 − 𝟑]

 t-statistics for sample excess kurtosis is as follows

𝐾𝑥 − 3
𝑡=
24
𝑇

𝐻0 : 𝐾𝑥 − 3 = 0

𝐻𝑎 : 𝐾𝑥 − 3 ≠ 0

Null hypothesis is rejected if p-value is less than significance level

(let’s say 0.05)
 Jarque-Berra (JB) Test
 Jarque Berra test combines the t-tests for skewness and excess kurtosis.

2 2
𝑆𝑥 𝐾𝑥 − 3
𝐽𝐵 𝑆𝑡𝑎𝑡𝑖𝑠𝑡𝑖𝑐 = +
6 24
𝑇 𝑇

𝑆𝑥2 𝐾𝑥 − 3 2
𝐽𝐵 𝑆𝑡𝑎𝑡𝑖𝑠𝑡𝑖𝑐 = +
6 24
𝑇 𝑇
𝐻0 : Underlying Time Series is normal in nature
𝐻𝑎 : Underlying Time Series is non-normal

Reject H0 if p-value is less than the significance level (let’s say 0.05 or 5%)
 Coefficient of Correlation
 Coefficient of correlation between two time series x and y is expressed
as follows.

𝜎𝑥,𝑦
𝜌𝑥,𝑦 =
𝜎𝑥 × 𝜎𝑦

1
σ𝑇𝑖=1 𝑥𝑖 − 𝜇𝑥 × (𝑦𝑖 − 𝜇𝑦 )
𝜌𝑥,𝑦 = 𝑇−1
1 𝑇 1 2
σ
× 𝑖=1 𝑥𝑖 − 𝜇𝑥 2 × σ𝑇𝑖=1 𝑦𝑖 − 𝜇𝑦
𝑇−1 𝑇−1

𝐸[ 𝑥 − 𝜇𝑥 × 𝑦 − 𝜇𝑦 ]
𝜌𝑥,𝑦 =
2 2
𝐸 𝑥 − 𝜇𝑥 × 𝐸 𝑦 − 𝜇𝑦
𝑤ℎ𝑒𝑟𝑒 − 1 ≤ 𝜌𝑥,𝑦 ≤ 1
 Coefficient of Correlation vs. Autocorrelation
Functions
σ𝑇𝑖=1 𝑥𝑖 − 𝜇𝑥 × (𝑦𝑖 − 𝜇𝑦 )
𝜌𝑥,𝑦 =
2
σ𝑇𝑖=1 𝑥𝑖 − 𝜇𝑥 2 σ𝑇𝑖=1 𝑦𝑖 − 𝜇𝑦

 Autocorrelation function depicts the linear dependence of a time series

(let’s say 𝑟𝑡 ) with its past values (𝑟𝑡−𝑖 ).

 ACF between 𝑟𝑡 and 𝑟𝑡−𝑙 is denoted as 𝜌𝑙

σ𝑇𝑡=2 𝑟𝑡 − 𝑟ҧ × (𝑟𝑡−1 − 𝑟)ҧ

𝜌1 =
σ𝑇𝑡=1 𝑟𝑡 − 𝑟ҧ 2
 Autocorrelation Function (ACF)

σ𝑇𝑡=3 𝑟𝑡 − 𝑟ҧ × (𝑟𝑡−2 − 𝑟)ҧ

𝜌2 =
σ𝑇𝑡=1 𝑟𝑡 − 𝑟ҧ 2

Generalizing the above framework

σ𝑇𝑡=𝑙+1 𝑟𝑡 − 𝑟ҧ × (𝑟𝑡−𝑙 − 𝑟)ҧ

𝜌𝑙 =
σ𝑇𝑡=1 𝑟𝑡 − 𝑟ҧ 2
 Autocorrelation Function (ACF)

 If 𝑟𝑡 is i.i.d., then 𝐸 𝑟𝑡2 < ∞ and 𝜌1 would have a mean of

zero and a variance of 1/T.

ෞ1
𝜌
 So, the test-statistic for testing 𝜌1 = 0 is = 𝑇×𝜌
ෞ1
1
𝑇

 H0 is rejected if p-value of t ratio is less than the significance level

(let’s say 0.05)
 Box-Pierce Test

 Box and Pierce (1970) proposed a joint auto-correlation test in

the following lines.

𝑄 ∗ 𝑚 = 𝑇. ෍ 𝜌𝑙2
𝑙=1

𝐻0 : 𝜌1 = 𝜌2 = ⋯ . = 𝜌𝑚 = 0

𝐻𝑎 : 𝜌𝑖 ≠ 0 𝑓𝑜𝑟 𝑎𝑛𝑦 𝑖 ∈ {1,2, … 𝑚)

 Ljung-Box test

 Ljung and Box (1978) modified the Box-Pierce test to increase the
power of the test for finite samples in the following lines.

𝑚
𝜌𝑙2
𝑄 𝑚 =𝑇 𝑇+2 ෍
𝑇−𝑙
𝑙=1