You are on page 1of 18

‭Exam EMFIN‬

‭Previous exams‬

‭VT2023‬
‭ 1: OLS matrix dimension‬
Q
‭Considering the following classical linear regression model, y=Xβ+u, where you‬
‭haveT=600 observations, the number of regressors, k=8 (including one intercept‬
‭and 7independent variables), what are the dimensions of Xβ and uu' ?‬

‭Linear regression model: y=XB+u‬


‭-‬ ‭y = dependent variable‬
‭-‬ ‭X = the matrix of the independent variable‬
‭-‬ ‭B = coefficient vector‬
‭-‬ ‭u = error term‬

‭ = 600 = number of rows in X‬


T
‭K = 8 = numbers of columns in X‬
‭Intercept = 1‬

‭ β‬
X
‭X = T x k matrix = 600 x 8‬
‭β = k x 1 column vector = 8 x 1‬
‭Xβ = 600 x 1‬

‭ u’‬
u
‭u = T x 1 column vector = 600 x 1‬
‭u’ = 1 x T = 1 x 600‬
‭uu’ = 600 x 600‬

‭Answer: 600x1 and 600x600‬

‭Obs if there were 2 intercepts → (k x 1)*2‬


‭ 2: OLS properties‬
Q
‭Which of the following is a correct description of the OLS estimator?‬

‭The OLS estimator is unbiased in both small sample and large sample.‬

‭also:‬
‭●‬ ‭OLS is BLUE‬
‭‬ O
● ‭ LS estimator is unbiased which means‬‭𝐸(‬ β‭‬) = β
‭●‬ ‭OLS estimator is unbiased even with a small sample‬

‭ 3: ML Estimator properties‬
Q
‭Which statement is the correct one describing the maximum likelihood‬
‭estimator?‬

‭It is a large sample required estimator, consistent and asymptotically efficient.‬

‭ 4: RSS expressions‬
Q
‭Which of the following are mathematical expressions of the residual sum of‬
‭squares?‬

‭1 and 4‬
‭Q5: Regression results interpretations‬

‭Liquidity = 8 - 0,8 * Volatility + 0,9 * No. of Trade‬‭i‬ ‭+ u‬‭i‬

‭Det ända som är viktigt att kolla på här är:‬


‭●‬ ‭H0: = not significant → meaning if we reject this (T.stat > T‬‭c‭)‬ it is significant!‬
‭●‬ ‭Interpret the formula‬
‭○‬ ‭For 1: vi ser att det är en positiv relation mellan liquidity och No. of trade‬‭i‬
‭pga‬‭“+”‬‭samt att genom rejecta H0 = significant‬
‭○‬ ‭For 2: vi ser att det är useful att använda volatility då den är med i‬
‭formeln. → så denna är fel!‬
‭○‬ ‭For 3: vi ser att det är en negativ relation mellan liquidity och volatility‬
‭pga “‬‭-‭”‬ , samt att genom att rejecta H0 = significant‬
‭○‬ ‭For 4: vi ser detta genom att rejecta H0 = significant‬

‭Alternatives:‬
‭●‬ ‭There is a statistically significant and positive relationship between‬
‭liquidity and No. ofTrades at the 10% significance level.‬
‭1.‬ ‭look at the coefficient for the number of trades which is 0,9‬
‭2.‬ ‭Since the coefficient has standard error (SE) 0,5 we compare this against‬
‭the critical value for the 10% significance level which is 1,65.‬
‭ ‬,‭9‬
0
‭0‬,‭5‬
= ‭1,‬ ‭8‬
‭ ,8 > 1,65 (which means that we reject, T‬‭c‬ ‭> Sign.level‬‭= reject).‬
1
‭Since we reject the null hypothesis this means that the coefficient for the‬
‭number of trades is statistically significant at the 10% level.‬

‭●‬ I‭ ncluding volatility in the regression model is not useful in helping explain‬
‭the variations inliquidity.‬
‭1.‬ ‭look at the coefficient for volatility, which is 0,8‬
‭2.‬ ‭since the coefficient has a standard error of 0,2 we compare it against the‬
‭critical value for the 10% significance level.‬
−‭0‬,‭8‬
‭0‬,‭2‬
=− ‭4‬

‭ > 1,65‬
4
‭We conclude that the coefficient for volatility is statistically significant at‬
‭the 10% level. Therefore, including volatility in the regression model‬
‭helps explain the variation in liquidity.‬
‭Therefore, this is not true!‬

‭●‬ T
‭ here is a statistically significant and negative relationship between‬
‭liquidity and volatility at the 1% significance level.‬
‭1.‬ ‭To determine if there's a statistically significant relationship between‬
‭liquidity and volatility, we look at the coefficient for volatility, which is‬
‭-0.8. Since the coefficient has a standard error of 0.2, we compare it‬
‭against the critical value for the 1% significance level, which is‬
‭approximately 2.58.‬
‭Since |(-0.8)/0.2| > 2.58, we can conclude that the coefficient for volatility‬
‭is statistically significant at the 1% level. Therefore, there is a statistically‬
s‭ ignificant relationship between liquidity and volatility. However, the‬
‭statement implies a negative relationship, which is true since the‬
‭coefficient for volatility is negative.‬

‭●‬ T
‭ he intercept is significantly different from zero at the 5% significance‬
‭level.‬
‭1.‬ ‭To determine if the intercept is significantly different from zero, we look‬
‭at the intercept coefficient, which is 8. Since the intercept has a standard‬
‭error of 4, we compare it against the critical value for the 5% significance‬
‭level, which is approximately 1.96.‬
‭2.‬ ‭Since |8/4| > 1.96, we can conclude that the intercept is statistically‬
‭significant at the 5% level. Therefore, this statement is true.‬

‭ 6: Why error term‬


Q
‭Why do we include a disturbance term in a regression model?‬
‭Det ända vi måste tänka på här är att error term påverkar dependend variable och inte‬
‭independent variable.‬

‭ ‬ S‭ ome determinants of the dependent variable may be omitted or left out.‬



‭●‬ ‭There may be random influences on the dependent variable which cannot be‬
‭modelled, e.g.a hurricane, a computer failure.‬
‭●‬ ‭There may be measurement errors of the dependent variables that cannot be‬
‭modelled.‬

‭Q7 Excel Beta estimation‬

‭ 8 - Error term definition‬


Q
‭The error term is the horizontal distance from the actual data point to the fitted‬
‭line.‬
‭= false‬
‭●‬ ‭tänk att punkterna visar vertiklat (upp å ner)‬
‭ he error term, also known as the residual, is the vertical distance from the actual data‬
T
‭point to the fitted line. It represents the deviation of the observed value from the predicted‬
‭value by the regression model.‬
‭Q9 - MLE and OLS‬
‭Please select the right statements describing the differences between OLS and‬
‭MLE.‬
‭●‬ ‭The error term is assumed to follow N(0,σ‬‭2‬‭) in OLS.‬‭But we can assume any‬
‭distributionfor the error term in MLE. - true‬
‭●‬ ‭MLE can be used to estimate both linear and non-linear regression models‬
‭whereas OLScan be used to estimate linear regression models only. - true‬

‭ 10 - CLRM assumption‬
Q
‭Which of the following are‬‭NOT TRUE‬‭in describing‬‭the assumptions for the‬
‭classical linear regressions, and the reasons why such assumptions are‬
‭necessary?‬

‭Classical linear regression‬

‭Assumptions:‬
‭1.‬ ‭Required for‬‭unbiasedness‬‭= the error(s) term has‬
‭zero mean(s).‬
‭2.‬ ‭Required for‬‭unbiasedness‬‭= there is no‬
‭releationsip between the error and corresponding x‬
‭variate.‬
‭3.‬ ‭Required for‬‭efficiency‬‭= The error term has a‬
‭constant covariance/constant variance and finite over all values of x‬‭t‬‭.‬
‭4.‬ ‭Required for‬‭efficiency‬‭= errors are statistically‬‭independent of one another‬
‭5.‬ ‭Required for‬‭parameter testing‬‭= The error term follows‬‭a normal distribution‬

‭True:‬
‭●‬ T ‭ he error term follows a normal distribution, this is required for‬‭parameter‬‭testing‬‭.‬
‭●‬ ‭The error term has a constant covariance, this is required for‬‭efficiency‬‭.‬
‭●‬ ‭The error term has a zero mean, this is required for‬‭unbiasedness‬‭.‬

‭False:‬
‭●‬ T ‭ here is no correlation between the error term and the independent variables, this i‬
‭required for efficiency.‬
‭●‬ ‭The error term is statistically independent of one another, this is required for‬
‭unbiasedness. → fel då detta rör efficency‬
‭Q11 Corr and regre‬

‭●‬ I‭ n correlation analysis, we treat the dependant variable and the independant variables‬
‭in acompletely symmetrical way.‬
‭●‬ ‭In regression analysis, we assume that the independent variables are non-stochastic,‬
‭orhave fixed values.‬

‭Q12 Consisnency‬

‭= True‬

‭ his statement is true because of the definition of a consistent estimator. In statistics, an‬
T
‭estimator is considered consistent if it converges in probability to the true parameter value as‬
‭the sample size increases indefinitely.‬

‭Q13 Hypothesis testing‬


‭ 14 DF understanding‬
Q
‭Suppose that you want to establish an appropriate univariate time series model for‬
‭yourdata series, y . And you want to perform a Dickey Fuller test for your time‬
‭series.Which of the following is TRUE in describing the Dickey Fuller test?‬

‭True:‬
‭●‬ I‭ f we reject the null hypothesis for the Dickey‬
‭Fuller test, we conclude that the time series is‬
‭stationary.‬

‭False:‬
‭●‬ T ‭ he Dickey Fuller test is more robust than the augmented Dickey Fuller test.‬
‭●‬ ‭The null hypothesis for the Dickey Fuller test is that Δy contains a unit root.‬
‭●‬ ‭We reject the null hypothesis if the Dickey Fuller test statistics is more negative or‬
‭morepositive than the DF critical value‬
‭○‬ ‭we reject if DF<T‬‭c‬

‭ 15 Stationary process‬
Q
‭Which of the following processes is NOT stationary?‬
‭-‬ ‭It’s not stationary since it’s a random walk with a drift.‬

‭Two types of non-stationary‬


‭1.‬ ‭Random walk model with drift‬

‭2.‬ ‭Trend-stationary process‬


‭-‬ ‭called so since it’s stationery around a linear trend.‬

‭-‬ ‭u‬‭t‬ ‭= white noice distrubance term‬

‭ 16 ARMA model‬
Q
‭An ARMA(0,0) process is equivalent to a white noise process.‬
‭= True‬

‭Q17 AR(p) or AR(p-1)‬

‭AR(5)‬

‭AR(p)‬
‭MA(q)‬

‭Q18 ADF regression results‬

‭‬ M
● ‭ odel 2‬
‭●‬ ‭Significance level = 1%‬

‭ DF t.stat = - 3,8954‬
A
‭T‬‭c‬ ‭= - 3,457‬

‭We reject if DF<T‬‭c‬

‭We reject the null hypothesis since, -3,8954 < -3,457. This means it’s‬‭stationary‬‭.‬

‭Q19 acf and pacf‬


‭MA(2)‬

‭ 20 CAPM H0‬
Q
‭What is the null hypothesis of the Wald test for emperical CAPM?‬

‭H0: a = 0. it means CAPM holds empirically.‬

‭Q21 CAPM Wald test‬

‭-‬ ‭Följ formeln‬


‭ 22 Likelihood ratio test emperical model‬
Q
‭In order to test the Sharpe-Lintner CAPM model, which of the following‬
‭regressionmodels do you use, if you want to use the Likelihood ratio test approach? R‬
‭denotes the excess return whereas r denotes for return.‬

‭Restricted och Unrestricted models:‬

‭LR Hypothesis testing‬

‭ = 10‬
N
‭lnL = -1835 (unrestricted)‬
‭lnL* = -1844 (restricted)‬

‭ 0: a = 0‬
H
‭Significance level = 5%‬

‭ 𝑅‬‭‬ = ‭‬ − ‭2(‬ ‭𝑙𝑛𝐿‬ *− ‭𝑙𝑛𝐿‬)


𝐿
‭LR = -2(-1844 - (-1835))‬
‭LR = 18‬

‭ ‬‭c‬ ‭= 18,31‬
T
‭Reject if LR > T‬‭c‬
‭I do not reject since LR < T‬‭c‬‭.‬

‭ ero beta CAPM‬


Z
‭Which of the following statements are TRUE describing the Black’s zero-beta CAPM,‬
‭ifyour data includes 200 portfolios?‬

‭True:‬
‭●‬ T ‭ his model can be tested by likelihood ratio test approach.‬
‭●‬ ‭The LR test statistics for testing Black's CAPM follows a χ2(199) distribution.‬

‭False:‬
‭●‬ T ‭ his model can be tested by Wald test approach.‬
‭●‬ ‭One underlying assumption is that investors can lend and borrow at the risk-free rate.‬
‭○‬ ‭no it works when there is no risk-free rate available.‬

‭ he Zero-Beta CAPM, proposed by Black, is a version of the Capital Asset Pricing Model‬
T
‭that works when there is no risk-free rate available.‬
‭●‬ ‭Without a risk-free asset, the model uses the market portfolio and its zero-covariance‬
‭portfolio. It calculates the expected return on any feasible portfolio or security by‬
‭considering its covariance with the market portfolio.‬
‭●‬ ‭Unlike traditional CAPM, the Zero-Beta CAPM does not rely on the market portfolio.‬
‭It identifies special portfolios that lie on the inefficient frontier.‬

‭●‬ T
‭ he model rewrites the traditional CAPM equation to express the expected return of a‬
‭portfolio or security as a function of its covariance with the market portfolio.‬
‭●‬ T
‭ o test the Zero-Beta CAPM, we examine whether the covariance or beta coefficients‬
‭are zero. Rejecting the null hypothesis suggests that Black's version of the CAPM‬
‭model is not valid.‬

-‭ ‬ 1‭ -B*Gamma+Beta* expected return on market‬


‭-‬ ‭γ = gamma‬
‭-‬ ‭where γ is the return of the zero-beta portfolio with the market(treated as an‬
‭unobserved quantity)‬
‭-‬ ‭ι is an(N×1)vector of ones‬

‭If we reject, this means that we reject Blacks version of the CAPM.‬

‭ APM versions‬
C
‭Which of the following are TRUE on tests of CAPM, where N is the number of‬
‭riskyassets?‬

‭True:‬
‭●‬ B
‭ oth Sharpe-Lintner CAPM model and Black's zero-beta CAPM model assumes‬
‭investorsare mean-variance optimizers.‬
‭○‬ ‭Both the Sharpe-Lintner CAPM and Black's zero-beta CAPM are based on the‬
‭assumption that investors aim to maximize their expected returns for a given‬
‭level of risk, which is a fundamental principle of mean-variance optimization.‬

‭ oth Wald test statistics and Likelihood ratio test statistics follow a X‬‭2‭(‬ N) distribution‬
‭●‬ B
‭fortesting Sharpe-Lintner CAPM.‬
‭○‬ ‭In testing the Sharpe-Lintner CAPM, both the Wald test statistics and the‬
‭Likelihood ratio test statistics asymptotically follow a chi-squared distribution‬
‭with N degrees of freedom, where N is the number of risky assets.‬

‭False:‬
‭●‬ B
‭ oth Sharpe-Lintner CAPM model and Black's zero-beta CAPM model have the‬
‭nullhypothesis: H: a = 0‬
‭○‬ ‭In the Sharpe-Lintner CAPM H0: B = 0‬
‭○‬ ‭In Black's zero-beta CAPM H0: a = 0‬
‭●‬ B
‭ oth Sharpe-Lintner CAPM model and Black's zero-beta CAPM model can be tested‬
‭byWald test approach.‬
‭○‬ ‭While the Wald test approach can be used to test hypotheses in statistical‬
‭models, it is not always applicable or optimal for testing all hypotheses in‬
‭financial models like the CAPM. While the Wald test can be used for some‬
‭hypotheses within the CAPM framework, other tests such as the Likelihood‬
‭ratio test are also commonly employed for different hypotheses. Additionally,‬
‭the use of specific tests may depend on the assumptions and structure of the‬
‭model being tested. Therefore, it's not accurate to claim that both CAPM‬
‭models can be tested solely by the Wald test approach.‬

‭ APM test stat‬


C
‭Both Wald test statistic and Likelihood statistic are always positive.‬
‭= true‬

‭Wald test stat‬

‭ ikelihood statistic‬
L
‭𝐿𝑅‬ =− ‭2‬(‭𝑙𝑛𝐿‬ *− ‭𝑙𝑛𝐿‬)

‭Event study H0‬

‭ ample = include 10 firms‬


S
‭It means that issuing new equity has no effects at all on firm values in general.‬

‭ he null hypothesis states that the average abnormal return (CAR) across all firms during the‬
T
‭subset event window from time s1 to time s2 is equal to zero. This implies that, on average,‬
‭there is no significant impact of issuing new equity on firm values across the entire sample of‬
‭firms. Therefore, the correct interpretation is that issuing new equity has no effects at all on‬
‭firm values in general.‬

‭ vent study assumptions‬


E
‭Choose the underlying assumptions for performing an event study analysis.‬

‭The underlying assumptions‬


‭●‬ ‭the market is efficient‬
‭●‬ ‭the event was unanticipated‬
‭●‬ ‭there were no confounding effects during the event window.‬
‭True:‬
‭●‬ T ‭ he event was a surprise to market participants.‬
‭●‬ ‭Financial market is efficient.‬
‭False:‬
‭●‬ ‭There can be more than one event occuring during the event window.‬
‭○‬ ‭only one‬
‭●‬ ‭The event was anticipated by some of the market participants.‬
‭○‬ ‭no unanticipated‬

‭ MFIN Event study CAR‬


E
‭Suppose you have performed an event study analysis, and estimated the‬
‭abnormalreturns for 6 securities and for the entire event window [-5, 5],‬

‭ vent study procedure‬


E
‭Which of the following statements are TRUE if you would like to do an event‬
‭studyanalysis to examine whether stock split affects firm values.‬
‭True:‬
‭●‬ ‭The expected normal returns can be estimated by several models, e.g. constant‬
‭meanmodels, market models, etc‬
‭○‬ ‭There are various models available to estimate expected normal returns,‬
‭including but not limited to constant mean models, market models (which‬
‭relate the returns of an asset to the returns of a market index), and more‬
‭sophisticated models like the Fama-French three-factor model.‬
‭●‬ T ‭ he parameters that used to calculate the expected normal returns for the event‬
‭windoware estimated over the estimation window.‬
‭○‬ ‭The parameters of the chosen model used to calculate expected normal returns‬
‭are typically estimated using data from the estimation window to ensure that‬
‭they capture the underlying relationship between the asset's returns and‬
‭relevant factors.‬
‭False:‬
‭●‬ ‭The estimation window and event window can overlap in some cases.‬
‭○‬ ‭The estimation window and event window should not overlap to ensure the‬
‭independence of the estimation of expected normal returns from the analysis‬
‭of abnormal returns during the event period.‬
‭●‬ ‭The abnormal returns can only be aggregated either across firms or through time.‬
‭○‬ ‭Abnormal returns can be aggregated across both firms and time. For example,‬
‭researchers may aggregate abnormal returns across firms to analyze the‬
‭average impact of an event across a sample of companies, or they may‬
‭aggregate abnormal returns over time to examine how the impact of the event‬
‭evolves over the event window.‬

‭Event study HT‬

‭T.test = Mean(Car)/S.e.(Car)‬

‭Event study general‬


‭Calculate test statistics for event studies‬

‭𝐶𝐴𝑅‬(‭𝑠‬‭1,‬ ‭𝑠‭2‬ ‬)

‭SD =‬ ‭𝑉𝑎𝑟‬(‭𝐶𝐴𝑅‬)

You might also like