You are on page 1of 40

FACULTAD DE

CIENCIAS
SOCIALES

Finanzas
Preparación para Certificaciones

Quantitative Methods (2):


READING 4 COMMON PROBABILITY
DISTRIBUTIONS
Diego Alonso Chicana Gutierrez, CFA, FRM
Based on Kaplan Schweser Material
Discrete and Continuous Distributions

• A probability distribution gives the probabilities of all possible


outcomes for a random variable.
• A discrete distribution has a finite number of possible outcomes.
• A continuous distribution has an infinite number of possible
outcomes.
• A probability function, p(x), specifies the probability that a
random variable equals a particular value, x.
• A cumulative density function (CDF): probability that a random
variable will take on a value less than or equal to a specific value.
e.g., p(x) = x / 10 for X = {1,2,3,4} → p(3) = 30%
→ CDF(3) = 60%
Uniform Distributions

A discrete uniform distribution has a finite number of possible


outcomes, all of which are equally likely.

Example: p(x) = 0.2 for X = {2,4,6,8,10}


p(2) = p(4) =p(6) = p(8) = p(10) = 0.2

A continuous uniform distribution: Probability distributed


evenly over an interval
Example: Random variable is continuous uniform over the interval 2 to 12.
P(X < 2) = 0 P(X > 12) = 0
P(4 ≤ X ≤ 8) = (8 – 4) / (12 – 2) =
2 3 4 5 6 7 8 9 10 11 12
4 / 10 = 40%
Binomial Random Variable

Each trial is a Bernoulli random variable; there are only two


possible outcomes. Probability of success on each trial (p) is
constant; trials are independent. A binomial random variable
gives us the probability of exactly x successes in n trials =
𝒄𝒏𝒙
𝐧!
𝑝 𝑥 = 𝑝𝑥 1 − 𝑝 𝑛−𝑥
𝐧 − 𝐱 ! 𝐱!

E(X) = np Var(X) = np(1 – p)


Properties of Normal Distribution

▪ Completely described by mean and variance


▪ Symmetric about the mean (skewness = 0)
▪ Kurtosis (a measure of peakedness) = 3
▪ Linear combination of normally distributed random variables is
also normally distributed
▪ Probabilities decrease further from the mean, but the tails go
on forever

Multivariate normal: more than one random variable, need


means, variances, and correlation coefficients
Confidence Interval: Normal Distribution

A range of values within which a random variable is expected to


be a certain percentage of the time.

68% confidence interval = X ± 1.00s


90% confidence interval = X ± 1.65s
95% confidence interval = X ± 1.96s
99% confidence interval = X ± 2.58s
Standard Normal Distribution
To standardize a random variable, calculate the z-value:
𝐱−𝒖 z is the number of standard
𝐳= ~𝑁 0,1 deviations from the mean.
𝝈

Example: The EPS for a large group of firms are normally distributed and
have μ = $6.00 and σ = $2.00. Find the probability that a randomly selected
firm’s earnings are greater than $9.70.
9.70 − 6.00
𝑧= = 1.85 $9.70 is 1.85 std. dev. above the mean of $6.00.
2.00 Cumulative Probabilities for a Std. Normal Distribution
Z 0.05 0.06
P(z ≤ 1.85) = 0.9678
1.7 0.9599 0.9608
P(z > 1.85) = 1 – 0.9678 1.8 0.9678 0.9686
= 0.0322 = 3.22% 1.9 0.9744 0.9750
Properties of Student’s t-Distribution
◼ Symmetrical (bell shaped)
◼ Fatter tails than a normal distribution
◼ Defined by single parameter, degrees of freedom (df), where
df = n – 1
◼ As df increase, t-distribution approaches normal distribution
Shortfall Risk and Safety-First Ratio

Shortfall risk: Probability that a portfolio return or value will be


below a target return or value
Roy’s safety first criterion Number of std. dev. target is below
the expected return/value
A portfolio with a higher SFRatio is preferred

𝐸 𝑅𝑝 − 𝑅𝐿
SF. 𝑅atio = , where RL threshold/target return
𝜎𝑝
Lognormal Distribution
• If x is normal, ⅇ𝑥 is lognormal
• Lognormal is always positive, used for
modeling asset prices
• Positively skewed
Continuous Compounding (CCR)
If we increase the number of compounding periods (n) for an
annual rate of return, the limit as n goes toward infinity is
continuous compounding. Example: For 18.23% stated
For a specific holding period return (HPR): annual rate with continuous
𝑒𝑛𝑑𝑖𝑛𝑔 𝑣𝑎𝑙𝑢𝑒 compounding. EAR?
𝐻𝑃𝑅 = − 1 = 𝑒 𝐶𝐶𝑅 − 1 EAR = ℮0.1823 – 1 = 20%
𝑏𝑒𝑔𝑖𝑛𝑛𝑖𝑛𝑔 𝑣𝑎𝑙𝑢𝑒
10 ℮0.1823 = 10 (1 + 20%)
Simulation

• Monte Carlo Simulation: Performed by making assumptions


about the distributions of prices or risk factors and using a large
number of computer-generated random values to generate a
distribution of possibly outcomes (e.g., project NPVs, portfolio
values).

• Historical Simulation: Similar to Monte Carlo simulation, but it


generates random variables from distributions of historical
data.
FACULTAD DE
CIENCIAS
SOCIALES

Finanzas
Preparación para Certificaciones

Quantitative Methods (2):


READING 5 SAMPLING AND ESTIMATION

Diego Alonso Chicana Gutierrez, CFA, FRM


Based on Kaplan Schweser Material
Sampling
To make inferences about the parameters of a population, we will use a
sample.
Sampling error is the difference between a sample statistic and true
population parameter (e.g., 𝑥ҧ − 𝜇).
1. Probability sampling refers to selecting a sample when we know the
probability of each sample member in the overall population.
2. Non-probability sampling use judgment of researcher, or lowcost/readily
available data, to select sample items. Less randomness in selection may
lead to greater sampling error.
Types of Sampling
1. Probability sampling
• Simple random: every population member has an equal probability of
being selected
• Stratified random: Separate the population into groups based on one or
more characteristics (1). Take a random sample from each class based
on the group size (2).
• Cluster: Create subsets (clusters), each of which is representative of an
overall population (e.g., personal incomes of residents by country).
2. Non-probability sampling
• Convenience: use readily available low cost data, not necessarily
representative, perhaps for preliminary investigation.
• Judgmental: select observations from population based on analyst’s
judgment.
Central Limit Theorem

For any population with mean μ and variance σ2

As the size of a random sample gets large, the distribution of


sample means approaches a normal distribution with mean μ
and variance σ2 /n.

Allows inferences about and confidence intervals for population


means based on sample means
Standard Error of the Sample Mean
Standard deviation of the distribution of sample means (sampling distribution)
for samples of size n.
𝝈 𝒔
𝝈𝒙ഥ = 𝒐𝒓 𝒔𝒙ഥ =
𝒏 𝒏

We use the standard error to construct a confidence interval (range of


values in which the population parameter is expected to lie).

Confidence interval for a population mean


= point estimate +/– (reliability factor) × standard error

For example: with a large sample size, a 95% confidence interval for the
population mean is the sample mean ±1.96 standard errors.
Confidence Intervals for Mean
When sampling from a: Reliability Factors
Small Sample Large Sample
Distribution Variance
(n < 30) (n > 30)
Normal Known z-statistic z-statistic
Normal Unknown t-statistic t-statistic*
Nonnormal Known Not available z-statistic
Nonnormal Unknown Not available t-statistic*

*The z-statistic is theoretically acceptable here, but use of


the t-statistic is more conservative.
Desirable Estimator Properties

Unbiased: expected value equal to parameter

Efficient: sampling distribution has smallest variance of


all unbiased estimators

Consistent: larger sample → better estimator


Types of Bias

• Data-mining bias – from repeatedly doing tests on same data


sample
• Sample selection bias – sample not really random
• Survivorship bias – sampling only surviving firms, mutual
funds, hedge funds
• Look-ahead bias – using information not available at the time
to construct sample
• Time-period bias – relationship exists only during the time
period of sample data
FACULTAD DE
CIENCIAS
SOCIALES

Finanzas
Preparación para Certificaciones

Quantitative Methods (2):


READING 6 HYPOTHESIS TESTING

Diego Alonso Chicana Gutierrez, CFA, FRM


Based on Kaplan Schweser Material
Steps in Hypothesis Testing

1. State the hypothesis—relation to be tested.


2. Select a test statistic.
3. Specify the level of significance.
4. State the decision rule for the hypothesis.
5. Collect the sample and calculate statistics.
6. Make a decision about the hypothesis.
7. Make a decision based on the test results.
Null and Alternative Hypotheses

Null hypothesis (H0) – contains =, ≥, or ≤


1. The hypothesis to be tested
2. Researcher “wants” to reject it
3. Always includes the equal sign
Alternative hypothesis (Ha)
1. What the researcher would like to conclude
2. What is concluded if the researcher rejects the null hypothesis
Two-Tailed Test
Use when testing to see if a population parameter is different
from a specified value
H0: μ = 0 5% significance level (probability in
both tails)
Ha: μ ≠ 0

Confidence
level
One-Tailed Tests
Use when testing to see if a parameter is above or below a
specified value.
H0: μ ≤ 0 vs Ha: μ > 0

H0: μ ≥ 0 vs Ha: μ < 0

significance
Confidence
level
level
Test Statistic and Critical Values
A test statistic is (1) calculated from sample data and (2) compared to critical
value(s) to test H0.
• The most common test statistics are the z-statistic and the t-statistic
• Which statistic you use to perform a hypothesis test will depend on the
properties of the population and the sample size as noted previously.

Critical values come from tables and are based on the researcher’s desired
level of significance. The critical values are obtained from the tables (t-
critic utilizes n-1 degrees of freedom).
If the test statistic exceeds the critical value (or is outside the range of critical
values), the researcher rejects H0.
𝑥ҧ − 𝜇0 𝑥ҧ − 𝜇0
z_𝑠𝑡atistic = σ 𝑡_𝑠𝑡atistic = 𝑠
𝑛 𝑛
Type I and Type II Errors
Type I Error: Rejecting H0 when it is actually true
Type II Error: Failing to reject H0 when it is false

Significance level is Probability of Type I Error [e.g., convicting


an innocent person (null is innocent)]
Power of a test is 1 – Probability of Type II Error (e.g., failing to
convict a guilty person)

Statistical significance does not necessarily imply economic


significance. Possible reasons: transactions costs, taxes, risk.
p-value

A p-value is the smallest level of significance at which the null


can be rejected, or the probability of getting the test statistic by
chance if the null is true.

If the p-value is given as 0.0214 or 2.14%:


We can reject the null at 5% significance.
We can reject the null at 3% significance.
We cannot reject the null at 1% significance.
Differences in Means, Mean Differences
Two tests of whether the difference between the means of two
populations equals a given value (often 0)
𝑥1ҧ − 𝑥ҧ2
• Difference in means test: 𝑡_stat =
Two independent normal populations. 𝑠𝑝2 𝑠𝑝2
+
Assume population variances are equal and use 𝑛1 𝑛2
pooled estimate of variance using both samples.
t-test, reject if test statistic is outside critical values
𝑑ҧ − 𝑢0
• Mean differences (paired comparison) test: 𝑡_statistic =
Two dependent normal populations 𝑆𝑑ത
Numerator is average difference between paired observations
Denominator is standard deviation of the differences t-test, reject if test
statistic is outside critical values
Other Hypothesis Tests

Nonparametric either do not consider a particular population parameter or


have few assumptions about the sampled population. (e.g., runs tests, rank
correlation tests).
FACULTAD DE
CIENCIAS
SOCIALES

Finanzas
Preparación para Certificaciones

Quantitative Methods (2):


READING 7 INTRODUCTION TO LINEAR
REGRESSION
Diego Alonso Chicana Gutierrez, CFA, FRM
Based on Kaplan Schweser Material
Linear Regression
Simple linear regression is used to estimate the linear relationship between
two variables and evaluate the significance of the relationship.

Minimizes the sum of the Rstock “Residual” or “error”


squared residuals term, actual minus
2% predicted = 0.8%
1%
Intercept 𝑏෠0 = 0.5%
(predicted value
Rmarket
-3% -2% -1% 1% 2% 3%
of Y when X = 0) -1%
Slope 𝑏෠1 = 0.9
Slope -2% (change in Y for 1 unit change in X)
Estimated regression equation
Yi = b0 + b1 Xi + Ɛi 𝑅෠𝑠𝑡𝑜𝑐𝑘 = 0.5% + 0.9𝑅෠𝑚𝑎𝑟𝑘𝑒𝑡
Regression Coefficients

cov 𝑋, 𝑌
𝑏෠1 =
𝜎𝑋2

𝑏෠0 = 𝑌ത − 𝑏෠1 𝑋ത Regression line passes


ത 𝑌)
through (𝑋, ത

OLS estimates – minimizes the sum of squared errors (SSE)


Assumptions of Linear Regression

1. Linear relation between dependent and independent variables


2. Variance of the error terms is constant (homoskedasticity)
3. Error terms are independently distributed (i.e., uncorrelated)
with each other
4. Error terms are normally distributed
Analysis of Variance (ANOVA)
Statistical procedure for analyzing the total variation in the
dependent variable.

SST

SSE + SSR = SST


ANOVA Table for Regression With n = 5
Source df Sum of Squares Mean Square
Regression 1 88.0 88.0
Residual (error) 3 7.2 2.4
Total 4 95.2

Degrees of freedom? Regression = 1, Residual = n – 2, Total = n - 1


Total sum of squares? = 88.0 + 7.2 = 95.2
Mean square regression (MSR)? = SSR / df = 88.0 / 1 = 88.0
Mean square error (MSE)? = SSE / df = 7.2 / 3 = 2.4
ANOVA Table for Regression With n = 5
Source df Sum of Squares Mean Square
Regression 1 88.0 88.0
Residual (error) 3 7.2 2.4
Total 4 95.2
𝑀𝑆𝑅 88.0
F-stat? = = 36.67
𝑀𝑆𝐸 2.4
𝑆𝑆𝑅 88.0
Coefficient of determination (R2)? = = 0.924
𝑆𝑆𝑇 95.2
Standard error of estimate (Se)? 𝑀𝑆𝐸 = 2.4 = 1.549

For simple regression, R2 = 𝒄𝒐𝒓𝒓𝒆𝒍𝒂𝒕𝒊𝒐𝒏𝟐𝒙𝒚


Regression Coefficient t-Test
Estimated slope (b1) = 0.81 with Sb1=0.086. Assuming n = 32,
determine if slope is equal to one at the at 5% significance level.
Test the hypothesis that b1 = 1.
0.81 − 1 Critical two-tailed t-stat (30 df
𝑡𝑏1 = = −2.209
0.086 and 5% significance) is 2.042
Reject H0: b1 = 1
For test of H0: b1 = 0, this is same as significance of F-stat
Predicted Values for “Y” Variable

Given b0 = 1.8% and b1 = 0.76 and a value of X of 8%, what is


the predicted value of Y according to this regression model?

𝑌෠ = 1.8% + 0.76 8% = 𝟕. 𝟖𝟖%


Confidence Interval for Predicted Y

Given a predicted value of Y of 7.88% and standard error of forecast


(sf) of 1.22% for a regression with 27 observations, construct a 95%
confidence interval (critical value with 25 df is 2.06).

7.88 ± 2.06 × 1.22 [5.37% , 10.39%]

1 𝑥 − 𝑥ҧ 2
𝑆𝑓2 = 𝑆𝐸𝐸 2 1+ +
𝑛 𝑛 − 1 𝑠𝑥2

sf can be approximated with SEE for large samples.


Functional Forms
When the relationship between X and Y is not linear, fitting a linear
model would result in biased predictions.
Transforming one or both variables by taking their natural logarithm
might make the relationship between the transformed variables linear.
Dependent Independent
Model Slope Interpretation
Variable Variable
Relative change in Y,
Ln (Y) X Log-lin absolute change in X
Forecast Y is elnY
Absolute change in Y,
Y Ln(X) Lin-log
relative change X
Relative change in Y,
Ln (Y) Ln(X) Log-log relative change X
Forecast Y is elnY

You might also like