lOMoARcPSD|5612673

Cheat sheet quantitative methods in finance nova

Quantitative Methods in Finance (Universidade Nova de Lisboa)

StuDocu is not sponsored or endorsed by any college or university

Downloaded by Bassirou Ndao (bassirou.ndao@gmail.com)
 lOMoARcPSD|5612673

R2 and SER OLS – Estimator

Standard error of the regression (SER) is just the standard deviation of the
residuals. R2 measures the fit relative to the variance of the dependent
variable, the SER just measures the fit. The rankings of the different dependent variables
(1st best, 2nd …) would only necessarily be the same reg. both R2
and SER if all the dependent var had the same variance
𝑅𝑆𝑆 + 𝐸𝑆𝑆 𝐸𝑆𝑆 𝑅𝑆𝑆 𝑅𝑆𝑆
𝑇𝑆𝑆 = 𝑅𝑆𝑆 + 𝐸𝑆𝑆 ↔ 1 = 𝑤𝑖𝑡ℎ 𝑅2 = = + 𝑅2 ↔ 𝑅2 = 1 −
𝑇𝑆𝑆 𝑇𝑆𝑆 𝑇𝑆𝑆 𝑇𝑆𝑆
R2 – Coefficient of determination (McFadden)
R2 cannot be compared when dependent variables are different (for e.g. Y≠logY)
R2 increases with the number of variables, even if the additional variables are not statistically
significant; therefore bigger R2 does not mean better, must compute the Adjusted - on average the estimators are the true values (unbiased): Ε(β ̂ )=β
- all formulae of the estimators are true linear combinations of random variables
𝑅̅2 (See Formula)
- variance of the coefficients is minimized (efficient)
Only compare LPM’s R2 with LPM’s R2; Logit’s R2 with Logit’s R2; Probit’s R2 with
- for an infinite number of observations, the estimators will converge to their true values (consistency)
Probit’s R2
𝑙𝑛𝐿𝑢
𝐹𝑜𝑟 𝒍𝒐𝒈𝒊𝒕 𝑎𝑛𝑑 𝒑𝒓𝒐𝒃𝒊𝒕: 𝑅 2 = GAUSS-MARKOV Assumptions/ Implications/ Test/ Solutions
𝑙𝑛𝐿0
- need to be fulfilled otherwise model (regression) might not be adequate/ efficient!
Test for significance (“Coefficient, constant = 0”)
I. Linearity: Errors have zero mean: Ε(𝑢𝑡 ) = 0
T-test: inferences about statistical significance regression coefficients (t-value)
II. Exogeneity: independent variable and error term are uncorrelated
Η0 : 𝛽𝑖 = 0, Η1 : 𝛽𝑖 ≠ 0, if t > critical value (mostly 1,96) then reject Η0 , value is significant Endogeneity: some of explanatory variables are correlated with the equations error term
Coefficient−Η0 IMPLICATIONS: OLS estimator biased and inconsistent
t-value = TEST: Durbin-Wu-Hausmann Test
Standard Error
SOLUTION: Instrumental Variables (IV) – when #instruments = #endogenous variables (exactly identified system);
Type I error is rejecting the null when it is true, Type II error is accepting the null when it is false. The probability of type I Or Two Stage Least Squares (2SLS) – when #instruments > #endogenous variables (overidentified system)
error is the significance level (for e.g. 5%).
III. Homoskedasticity: variance of errors is constant and finite
Indication p-value:
Heteroskedasticity: variance of the errors is not constant across observations
Low value: < 0.05 - high evidence against Ho – reject = coefficient significant
IMPLICATIONS: Estimators are still unbiased, and consistent (OLS not violated); however, no longer efficient - (not
High value: > 0.05-0.1 few evidence again Ho = not significance
min. variance); hence t-& F-test no longer reliable, thus possibly erroneous inferences regarding stat significance and wrong
Confidence interval signifiance: standard errors since error terms no longer normally distributed
Η0 : 𝐶𝑜𝑛𝑠𝑡𝑎𝑛𝑡 𝑜𝑟 𝛽𝑖 = 0, Η1 : 𝐶𝑜𝑛𝑠𝑡𝑎𝑛𝑡 𝑜𝑟 𝛽𝑖 ≠ 0 TEST: White’s Test
SOLUTION: Huber-White Correction “Robust estimation”/
If 0 /∈ CF (not part of the CF) then reject Η0
“Adjust: Robust Standard errors”
F-test: Overall significance test of unrestricted regression
𝛽̂ − 𝛽0 𝛽̂ − 𝛽0
𝑡= =
𝑆𝐸
̂ 𝟐𝒊
∑ (𝑥𝑖 − 𝑥̅ )2 × 𝒖
√
K = #regressors (including constant); T = #observations [∑ (𝑦𝑖 − 𝑦̅) ]𝟐
2
Test of significance of overall model (F-value) = High value = Signifiance
IV. No Autocorrelation: no pattern in the errors (residuals)
F-test: Joint, overall significance test (test of the validity of the restrictions) Autocorrelation: pattern in the residuals (assumes that relationship is between an error and previous one)
Ϝ𝛼𝑟, 𝑇−𝐾−1 ∼ CV IMPLICATIONS: Estimators are no longer BLUE
Static model (y=ax1t+bx2t)-inefficient estimators
(𝑅𝑅𝑆𝑆−𝑈𝑅𝑆𝑆)/𝑟 Dynamic model (y=ax1t+bx2t-1)- inconsistent estimators (have to change my model)
Η0: 𝛽1 = 𝛽𝑗 = 0 Η1: 𝛽1 ≠ 0, … .∨ 𝛽𝑗 ≠ 0 𝐹 − 𝑇𝑒𝑠𝑡 =
𝑈𝑅𝑆𝑆/(𝑇−𝐾−1) TEST: Durbin Watson (DW-stat bounds) / Breusch-Godfrey Test (LM)
unrestricted: model contains all variables exactly as in overall regression SOLUTION: Newey-West’s HAC robust standard errors – Static Models
restricted: all regressors which coefficients have been set to “0” are excluded
degrees of freedom: (v2) N / T (sample size) – k (numbers parameters in unrestricted model)
and Multicollinearity: explanatory variables are very highly correlated with each other
(hints: high R2 and high SEs)
numbers of restrictions: (v1) r IMPLICATIONS:
̂
̂ −Ζ ̂
̂ ̂ ̂ ̂
̂ TEST: Variance Inflation Factors
Confidence Interval: [𝛽 1−𝛼/2 × 𝑆𝐸 (𝛽) ; 𝛽 + Ζ1−𝛼/2 × 𝑆𝐸(𝛽)] SOLUTION: drop collinear variables/ regress one on another and take residuals/ transform into ratio
Significance Tests
Η0 : 𝛽𝑖 = 0 Η1 : 𝛽𝑖 < 0 Η0 : 𝛽𝑖 = 0 Η1 : 𝛽𝑖 > 0 Η0 : 𝛽𝑖 = 1 Η1 : 𝛽𝑖 ≠ 1 Η0 : 𝛽𝑖 = 𝛽𝑗 Η1 : 𝛽𝑖 < 𝛽𝑗
→ negative correlation between x 1 and y → positive correlation between x 1 and y → one-to-one relationship → difference in impact on y

Coefficient Interpretation:
Level-Level: If x (independent variable) varies by 1 unit, y (dependent variable) varies by ß1 units on average ceteris paribus// Constant: If x1 (𝐶𝑜𝑒𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑡𝑠) = x2 = x3 = 0 then constant is the XYZ (e.g. excess return)
Level-Log: If x varies by 1%, y varies ß1/100 units on average ceteris paribus.
Log-Level: If x varies by 1 unit, y varies 100ß1% on average, ceteris paribus.
Log-Log: If x varies by 1%, y varies ß1% on average ceteris paribus.
Logit/ Probit: If 𝑥𝑖𝑗 (if 𝛽𝑗 > 0) increases by 1 unit, the probability of P(y=1) increases by 𝛽𝑗 , vice versa.

Downloaded by Bassirou Ndao (bassirou.ndao@gmail.com)

 lOMoARcPSD|5612673

OLS: Endogeneity (𝑐𝑜𝑣(𝑥𝑡 , 𝜐𝑡 ) = 0)

OLS: Heteroskedasticity OLS: Autocorrelation D-W-HAUSMAN TEST
WHITE’s (tests if unequal var in the error DURBIN-WATSON (Autocorr. of 1st order, regress residuals) Steps:
term) It assumes that the relationship is between an error and the previous 1.) Estimate original regression:
One: 𝑢𝑡 = 𝝆𝑢𝑡−1 + 𝑣𝑡 e.g. 𝑌𝑡 = 𝛽0 + 𝛽1 𝑥1𝑡 + 𝛽2 𝑥2𝑡 + 𝛽3 𝑥3𝑡 + 𝑣𝑡 add: “by OLS save residuals 𝑣̂𝑡 ”
How to detect: Dependent variable=Residual2 ∑𝑇
𝑡=2(𝑢 ̂𝑡−1 )2
̂𝑡 −𝑢
from original regression =auxiliary regression, DW Score formula: 𝐷𝑊 = ∑𝑇 ̂𝑡 2
𝑡=2 𝑢 2.) Estimate regression with instruments for the potential endogenous var.
Cross-multiples’ of all variables are added as
Test actually tests: 𝑯𝟎 : 𝝆 = 𝟎 and 𝑯𝟏 : 𝝆 ≠ 𝟎 e.g. 𝑥1𝑡 = 𝛼0 + 𝛾1 𝑧1𝑡 + 𝛾2 𝑧2𝑡 + 𝛾3 𝑧3𝑡 + 𝑢𝑡 add: “by OLS and save residuals 𝑢
̂𝑡 ”
regressors
Look into stat tables for 𝑛 and 𝑘, (!) 𝑘 not including the constant
Table: E_MKT/E_MKT^2/E_MKT*HML/etc. 3.) Regress first residuals from step 1.)
e.g. 𝑣̂𝑡 = 𝛿0 + 𝛿1 𝑥1𝑡 + 𝛿2 𝑥2𝑡 + 𝛿3 𝑥3𝑡 + 𝛿4 𝑢
̂ 𝑡 + 𝜀𝑡 add: “by OLS”
𝑯𝟎 : 𝜸𝟏 = ⋯ = 𝜸𝒙 = 𝟎 | 𝒉𝒐𝒎𝒐𝒔𝒌. (x = No. of
coefficients w/o constant) 4.) Test with if coefficient from step 3.) is significant:
𝑯𝟏 : 𝜸𝟏 ≠ ∨ … ∨ 𝜸𝒙 ≠ 𝟎 | 𝒉𝒆𝒕𝒆𝒓𝒐𝒔𝒌. 𝑯𝟎 : 𝜹𝟒 = 𝟎 | 𝒏𝒐 𝒆𝒏𝒅𝒐𝒈𝒆𝒏𝒆𝒊𝒕𝒚
(See example auxiliary regression = unrestricted 𝑯𝟏 : 𝜹𝟒 ≠ 𝟎 | 𝒆𝒏𝒅𝒐𝒈𝒆𝒏𝒆𝒊𝒕𝒚
below)
LM = 𝑇 ∗ 𝑅2 , Compare with χ𝐶𝐹−𝑖𝑛𝑡𝑒𝑟𝑣𝑎𝑙
𝑟 , if LM-value > χ then reject 𝐻0 , (!) 𝑅2 from regression of first residuals (Step 3), 𝑟
Then do number of endogenous variables (here just 1)
LM = 𝑇 ∗ 𝑅2 , Compare with χ𝐶𝐹−𝑖𝑛𝑡𝑒𝑟𝑣𝑎𝑙
𝑟 , if LM-
value > χ then reject 𝐻0 Instrumental Variables (To use when endogeneity was detected, exactly identified system)
Rejecting 𝐻0 = Evidence for Heteroskedasticity, 𝑧𝑡 =IV: Has to be (1) strongly correlated with endogenous variable & (2) uncorrelated with error term
Adjust w/ Robust SE If 𝑧𝑡 would be directly added, regression would be changed, Solution: first order derivation
IV Estimator:
Closest alternative test: ∑(𝑦𝑡 − 𝑦̅)(𝑧𝑡 − 𝑧̅)
Closest Alternative test is the regression
Limitations of test: 1.) Only testing 1st order autocorrelation 2.) not valid
𝛽̂
𝐼𝑉 =
for in dynamic models or under endogeneity ∑(𝑥𝑡 − 𝑥̅ ) (𝑧𝑡 − 𝑧̅)
significance F test provided in the auxiliary
2-Stage Least Squares (To use when endogeneity was detected, overidentified system)
regression output (look at the p-value of the BREUSCH-GODFREY (Autocorr. up to 4th order) 1.) Estimate regression with instruments for the potential endogenous var.:
overall significance). These are just the
hypothesis associated (only need them if they
How to detect: Dependent variable=Residual from original regression, e.g. 𝑥1𝑡 = 𝛼0 + 𝛾1 𝑧1𝑡 + 𝛾2 𝑧2𝑡 + 𝛾3 𝑧3𝑡 + 𝑢𝑡 add: “by OLS and save fitted values”
Lagged variables added to regression Note: Instruments are/ must be:
asked you to write them)
1.) Strongly correlated with endogenous variable, 2.) Uncorrelated with the error term, 3.) Not necessary for the model
Unrestricted: 𝑢 ̂ 𝑡2 = 𝛾0 + 𝜸𝟏 𝒙𝟏𝒕 + 𝜸𝟐 𝒙𝟐𝒕 + Table: [lag1resid]/ [resid-1] B-G
2.) Estimate regression of original model BUT with fitted values for endogenous variable:
𝜸𝟑 𝒙𝟐𝟏𝒕 + 𝜸𝟒 𝒙𝟐𝟐𝒕 + 𝜸𝟓 𝒙𝟏𝒕 𝒙𝟐𝒕 + 𝑣𝑡
𝑯𝟎 : 𝜸𝒙 = ⋯ = 𝜸𝒚 = 𝟎 | 𝒏𝒐 𝒂𝒖𝒕𝒐𝒄𝒐𝒓𝒓. (x, y = Range of lagged 𝑌𝑡 = 𝜗0 + 𝜗1 𝑥̂1𝑡 + 𝜗2 𝑥2𝑡 + 𝜗3 𝑥3𝑡 + 𝑣𝑡
̂ 𝑡2 = 𝛾0 + 𝑣𝑡
Restricted: 𝑢 Note: 1.) Coefficients are differently noted as they are different from first regression, 2.) Running 2-Stage OLS on a model
variables from the output)
w/o endogeneity makes it less efficient
𝑯𝟏 : 𝜸𝒙 ≠ ∨ … ∨ 𝜸𝒚 ≠ 𝟎 | 𝒂𝒖𝒕𝒐𝒄𝒐𝒓𝒓.
Binary dependent variables Panels 1 and 2: 2-stage SLS
Need to transform the dichotomous (zweigeteilt) Y LM = 𝑇 ∗ 𝑅2 , Compare with χ𝐶𝐹−𝑖𝑛𝑡𝑒𝑟𝑣𝑎𝑙
𝑟 , if LM-value > χ then reject 𝐻0 approach. In the 1st panel the
into continuous variable 𝑌′ ∈ (−∞, ∞) Solution Rejecting 𝐻0 = Evidence for 4th order autocorrelation, (!) 𝑟 just the endogenous variable is
= Link function 𝐹(𝑌) that takes a dichotomous 𝑌 number of lagged variables and 𝑇 = number of observations reduced by regressed on a set of IVs while in
and gives us a continuous, real-valued 𝑌′; lags, in plots already done (!) the 2nd stage the dep var is
Probit Model: Choice of # lagged error terms (monthly: 11, quarterly: 3, annually: 1) regressed on the fitted values
Which function does that? Cumulative normal from the 1st stage.
distribution 𝛷 , given any Z-score it gives =
𝛷(𝑍) ∈ [0,1]// It follows: 𝐹(𝑌) = 𝛷−1 (𝑌)
Linear (OLS) Non-Linear Approach (Formulas) Non-Linear Approach (Applications) Joint-tests
In a Probit model 𝛽𝑥𝑖 is taken to be the z-value of
a normal distribution 𝑦′𝑖 = 𝛼 + 𝛽𝑥𝑖 + 𝑢𝑡 1.) Express marginal effect (formula e.g. for unempl.) F-test =- not valid under Probit or Logit since
𝜕 𝑃(𝑦=1) 𝜕 𝐹(𝑥𝑖 ′𝛽)
Logit Model: P [y=1] = 𝑥𝑖 ′ ∗ 𝛽 Logit (Logistic distribution): 𝑀𝐸 𝑜𝑓 𝛽𝑢𝑛𝑒𝑚𝑝𝑙 = = 𝜕 (𝛽 = 𝑓(𝑥𝑖′ 𝛽) ∗ non-linear!
Cumulative distribution: 𝜕 (𝛽𝑢𝑛𝑒𝑚𝑝𝑙 ) 𝑢𝑛𝑒𝑚𝑝𝑙 )
Based on the odds ratio (Chancenverhältnis): Overall likelihood ratio-test > equivalent F-test
- derive for 𝑥𝑖 yields 𝛽𝑖 1 1 1
−2( 𝑥𝑖 ′𝛽)2
𝑂𝑅(𝑝) = 𝑝/(1 − 𝑝); Taking the log: 𝑙𝑜𝑔𝑖𝑡(𝑌) =
𝐹(𝑥𝑖 ′𝛽) = 𝛽̂𝑢𝑛𝑒𝑚𝑝𝑙 = ̂
𝛽2 ∗ 𝑒 Η0 : 𝛽1 = 𝛽𝑗 = 0 𝒏𝒐 𝒐𝒗𝒆𝒓𝒂𝒍𝒍 𝒔𝒊𝒈𝒏𝒊𝒇𝒊𝒄𝒂𝒏𝒄𝒆
- not useful in extremes √2𝜋
𝑙𝑜𝑔[𝑂(𝑌)] = 𝑙𝑜𝑔[𝑦/(1 − 𝑦)] 1 + 𝑒 −𝑥𝑖 ′∗𝛽
(negative results) 2.) Calculate an exact marginal effect Η1 : 𝛽1 ≠ 0 ∨ … .∨ 𝛽𝑗 ≠ 0 𝑜𝑣𝑒𝑟𝑎𝑙𝑙 𝑠𝑖𝑔𝑛𝑖𝑓𝑖𝑐𝑎𝑛𝑐𝑒
Density distribution: (use for marginal effect)
Properties: 𝑦“𝑖 𝑥𝑖 ′𝛽 𝑒 (𝑐𝑜𝑚𝑝𝑙. 𝑙𝑜𝑔𝑖𝑡 𝑚𝑜𝑑𝑒𝑙∗𝑚𝑒𝑎𝑛 𝑜𝑓 𝑠𝑎𝑚𝑝𝑙𝑒)
𝑒 1 𝑀𝐸𝑙𝑜𝑔𝑖𝑡 = 𝛽1 ∗ LR = - 2 [Lu – Lr] (likelihood unrestricted – restricted)
• always a constant in model = 𝛼 + 𝛽 𝑥𝑖 (𝐺𝑒𝑛𝑑𝑒𝑟) 𝑓(𝑥𝑖 ′𝛽) = 𝛽 ∗ → 𝐹(𝑡) = (1 + 𝑒 𝑐𝑜𝑚𝑝𝑙. 𝑙𝑜𝑔𝑖𝑡 𝑚𝑜𝑑𝑒𝑙 )2
(1 + 𝑒 𝑥𝑖 ′𝛽 )2 1 + 𝑒 −(𝑥𝑖 ′𝛽) ~ 𝜒2(𝑘) (number of restrictions ß)
• Certain function F gives outcomes only 0;1 P [y=1] | x (Gender) = 0] ME Probit: If overall likelihood test score > 𝜒 2, reject 𝐻𝑜 , Assume
(failure; success) – no negative probability →P=±𝛼 Probit (Probability unit): a.) Insert sample mean into probit equation, solve. overall significance
• need threshold: if <t favour zero 0; if >t Cumulative Distribution: 1 1 2
favour one 1 Non-linear approach
𝑦
𝐹(𝑦) = ∫−∞ ∅(𝑢) Integral from −∞ to 𝑦 of the normal PDF b.) 𝑀𝐸𝑝𝑟𝑜𝑏𝑖𝑡 = 𝛽1 ∗ 𝑒 −2(𝑟𝑒𝑠𝑢𝑙𝑡 𝑚𝑒𝑎𝑛 𝑖𝑛 𝑝𝑟𝑜𝑏𝑖𝑡 𝑒𝑞𝑢𝑎𝑡𝑖𝑜𝑛)
√2𝜋
• prone to Heteroskedasticity (different (𝑢 as dummy to integrate over). =Integrating up to the point Goodness of fit (for both, logit and probit):
P [y=1] = 𝐹(𝑥𝑖 ′ ∗ 𝛽) 3.) Calculate concrete value of 𝒙𝒊 for a known level of (P) 𝐿𝑜𝑔(𝐿 )
behaviour of error (u) with 0;1) 𝑦 Logit: 𝑅2 = 1 − 𝐿𝑜𝑔(𝐿1,𝑢𝑛𝑠𝑡𝑟) Note: R2 cannot be used to
- derive for 𝑥𝑖 yields Density function: (use for marginal effect) a.) Set known P level equal with logit cumulative function: 0,𝑟𝑒𝑠𝑡𝑟
chose between Logit/Probit as these two have different
𝑓(𝑥𝑖 ′ ∗ 1 1 2 1
𝑓(𝑥𝑖 ′𝛽) = 𝛽 ∗ 𝑒 −2𝑥𝑖 ′𝛽 → 𝐹(𝑥𝑖 ′𝛽) → 𝜙 𝑇𝑎𝑏𝑙𝑒𝑠! (z) 𝐿𝑜𝑔𝑖𝑡: 𝑃 𝑙𝑒𝑣𝑒𝑙 = likelihoods
𝛽 ) ∗ 𝛽𝑖 √2𝜋 1 + 𝑒−(𝐿𝑜𝑔𝑖𝑡 𝑓𝑢𝑛𝑐𝑡𝑖𝑜𝑛 𝑤𝑖𝑡ℎ 𝑥𝑖)
𝟏 b.) Solve for 𝑥𝑖
- weigh beta by density P [y=1] | x(Gender)=0] → F(𝑥𝑖 ′ ∗ 𝛽) = 𝐹(𝑦̂𝒊“) =
𝟏+𝒆−𝑥𝑖 ′∗𝛽 Probit:
function logit density
a.) To be solved: 𝛷(𝑝𝑟𝑜𝑏𝑖𝑡 𝑓𝑢𝑛𝑐𝑡𝑖𝑜𝑛) = 𝑃 𝑙𝑒𝑣𝑒𝑙
- strength of probability: b.) Short-cut: Look for Z-Score in Stat Table for P level and
Logit set:
c.) 𝑃𝑟𝑜𝑏𝑖𝑡 𝑓𝑢𝑛𝑐𝑡𝑖𝑜𝑛 = 𝑂𝑝𝑡𝑎𝑖𝑛𝑒𝑑 𝑍 − 𝑆𝑐𝑜𝑟𝑒 (Solve for 𝑥𝑖 )
Downloaded by Bassirou Ndao (bassirou.ndao@gmail.com)