Mock Exam A

What is the difference between agglomerative and divisive clustering?

Divisive clustering starts with a single population (top-down) and divides into many, meaning it is suitable for large populations. Agglomerative clustering focuses on local patters (bottom-up) and is better suited to small clusters

Which of agglomerative and divisive clustering is top-down/ bottom-up?

Divisive = Top-down (think dividing population into segments),

Agglomerative = Bottom-Up

How do you bypass the limitation of k-means clustering?

k-means clustering requires a hyperparameter (k), which is the number of clusters you intend to use. By knowing this beforehand, you bypass the limitation by setting how many ‘centroids’ will be used.

What hyperparamter is used to change the magnitude of the weighting adjustments made by a DNN as it processes?

Learning rate

What ML model does the hyperparamter ‘minimum population at a node’ relate to?

Classification and regression trees (CARTs)

What ML model does the hyperparamter ‘lamba’ relate to?

LASSO model

Other than the learning rate, what are the other hyperparamters of DNN models?

Number of hidden layers, nodes per hidden layer, and architecture of the connectivity and activation functions

What supervised ML algorithim is characterised as a penalised regression with the goal of removing features? What does this achieve?

LASSO is a penalised regression that removes features that explain little of the variation within the data. This achieves a reduction in overfitting, by only adding features that decrease the sum of squared residuals (SSR) by more than the penalty term (Lambda) increases.

What supervised machine learning algorithm would be most appropriate to reduce overfitting in DNN?

LASSO

What are we looking for with Area Under Curve (AUC) results?

Ideally similar and high values for both the training set, and cross-validation (CV) set, as well as the p-value

What does a higher training set and a lower Cross-Validation set AUC indicate?

Overfitting - model performs well on the data but less in the real world

What does a lower training set and a higher Cross-Validation set AUC indicate?

Underfitting, model performs well in real world but not as good with the training example

What does a good-fit indicate relevant for AUC results?

A high and similar training and CV score, and a high p-score

If the current trailing P/E is less than the justified trailing P/E, what does that mean?

The stock is undervalued

How to calculate justified trailing P/E

One-Period GGM

What’s Return on Equity (ROE), as a function of Return on Asset (ROA)?

ROA x Financial Leverage (equity multiplier)

What does overage rent allow property owners to do?

Take advantage of periods of strong sales and reduce rent payments when sales are not as strong

How do you calculate Overage rent rate required?

Overage rent (Suggested Rent Change) / Overage Sales

Why is MVaR useful as a pre-trade evaluation tool?

Because it explains the total impact of a trade on total VaR. The total risk budget equals the proportionally weighted sum of the individual MVaRs

What selection bias is related to backtesting strategies until a desired outcome is achieved?

Data Snooping/ p-hacking

How do you correct for suvivorship bias?

Point-in-time data, rather than only using current index constituents

When would credit spreads likely be narrow?

During periods of sustained economic expations and inflationary pressures

At what threshold does the beneish model indicate manipulation?

M-score higher than -1.78

What threshold indicates that shareownership is dispersed?

no one (group) controls more than 50% of the shares

What is dividend recapitalisation?

A form of balance sheet restructuring where the company issues low-cost debt and distributes the cash through dividends or the repurchasing of shares. This would increase the debt-to-EBITA ratio

When I do those stupid two period binominal discount model things for dividends, how do I remember what I need and don’t need to use?

Only rate needed is the furtherest right (t = 2) and only discount p-value I don’t need is the furthest right (it will be something like 0.9669). Stupid questions. I also only use the given interest rate at the start and assume .5 movement probabilities)

What are the requirements when a violation occurs under Standard IV(C) Responsibilities of a Supervisor?

• Respond promptly

• Conduct a thorough investigation

• Increase supervision and place limitations while the investigation is still ongoing, and

• review procedures to prevent future violations

The offenders explanation is not sufficiently cover conducting a throrough investigation, and a warning is not a sufficient limitation to ensure it does not occur again.

Under the standards, what are the requirements for personal trading?

• Do not harm clients by their trade,

• Do not personally benefit from the trades, and

• Adhere to governing laws and regulations

How to remember surprise formula

If the outcome is worse then the forecast then it follows that that would be negative (Actual - Forecast)

What is the most significant source of tracking error?

Difference in portfolio composition relative to index, i.e. the difference in portfolio composition because it is impractical to fully replicate every holding exactly due to cost considerations

What does a Transfer Coefficient of 0 indicate?

No relationship between active forecasts and active weights, and therefore no indication of skill, and the full fundamental law would = 0

What do you get by squaring TC? (TC²)

the active return attributed to skill, and 1 - TC² is attributed to constrained induced noise

What is an appropriate multiple to compare capital-intensive firms?

EV/EBITDA - because it is calculated before depreciation and amortization expenses are deducted

When is it appropriate to use the harmonic/ weighted harmonic mean?

As a central measure tendency, usually for things like portfolios when extreme values exist. Weighted harmonic mean is used when the portfolio is not equally weighted

When is it appropriate to use the winsorized mean?

As a central measure indicator, when clear outliers exist as it limits the top/bottom 5% impact

What should I do when calculating IR for full fundamental law?

Check rebalancing frequency, i.e. if it changed from annual to quarterly, the breadth should be x 4, before being square rooted

As interest rates fall below that of the coupon rates, what happens to the effective duration for option-free, putable, and callable bonds?

EffDur will increase for straight and putable bonds, but decrease for callable bonds

Bonds with similar maturity dates, call dates, etc. but with different Option-adjusted spreads (OAS), is the one with the lower OAS under or overvalued relative to the similar bond with a higher OAS?

Smaller OAS relative to bond with similar characteristics is overvalued, because it implies more cost per unit of par value

What is the impact on an embedded options OAS when they experience a decrease in volatility?

OAS will increase with a decrease in volatility

When bootstrapping the Fixed Income questions, what should I remember to do?

Don’t round early - go to at least 4 decimal places

What’s the trick with bootstrapping FRN notes?

FRN will give a floor or ceiling on the coupon. In the case of a cap/ceiling, this means that the added coupon does not exceed this, and vice versa for floors. In general you use the given tree rates as the coupons instead of the coupon itself, and when they exceed the floor/ceiling, use the FRN rate instead.

For embedded option bonds, a decrease in interest rate volatility is good and bad for:

Good for callables, bad for putables

What does a bonds price, compared to it’s conversion price tell us about a bond?

When the bond price is lower than the conversion price, it is it out of the money, and therefore will behave bondlike

When the conversion price is lower, and its value will reflect the number of shares obtained at the conversion value

how does the convexity of putable and callable bonds react to being out of the money?

Both will exhibit positive convexity

Which bonds exhibit positive and negative convexity?

Straight, putable and callable can exhibit positive convexity, but callable bonds can also exhibit negative convexity

Is the goodwill more using the full goodwill acquisition method, or partial? How does this impact ROE?

Full goodwill method will come out with more total goodwill from a transaction meaning the ROE will be smaller (As a larger Equity is used in the denominator)

How to calculate P/S?

Price / (Sales/Shares Outstanding)

Which is least subject to manipulation? Why? P/B, P/S, P/E

P/S

What is P/B most appropriate to value?

Companeis with mostly liquid assets, i.e. are not going concerns (i.e. liquidation value)

How to calculate the E component of P/E when using the average ROE method?

EPS = Average ROE * BVPS

How do I work out what Book Value is?

BV = Shareholders’ Equity - Preferred Equity, or

Total Assets - Total Liabilities - Preferred Stock

How do calculate Earnings Yield?

Earnings Yield is simply E/P (Reverse of P/E)

How to compute cashflow yield?

(CFO / Shares Outstanding) / Price

How to compute Cash Flow?

Earnings (Net Income) + Depreciation and Amortization

How do we determine which SUE value is best?

The absolute value closest to 0

What should you check for when comparing companies of different sizes? (QM)

Variable scaling

What should you check for when a relationship is logarithmic? (QM)

Inappropriate variable form

What does a VIF of 10 + indicate?

Indicates serious multicollinearity

What does a VIF of 5 + indicate?

Warrants further investigation of the variable for multicollinearity

is a high leverage point or outlier the one that sits above or below?

Outlier sits above or below (because it’s within the dataset still)

How do you determine if a datapoint is considered influential with leverage?

If the datapoints leverage exceeds leverage formula, then it is considered influential

3((k + 1) / n), where n = degrees of freedom

How to identify a slope dummy?

It mixes two variables i.e. (AUM_RISK)

How to identify the base category in a model?

It is omitted from the regression

What’s another name for LASSO?

Penalised regression

When is a Support Vector Machine (SVM) most appropriate?

When binary classification is needed

When is k-nearest neighbour most appropriate?

Classify new datapoints

What ML algorithm avoids using hyperparameters?

SVM

What’s a key component of Reinforcement Learning, compared to supervised/unsupervised learning?

Reinforcement Learning requires a reward they will aim to maximise

Formula for Loss Given Default (LGD)?

LGD = Expected Exposure x (1 - Recovery Rate)

What does a reduced-form model do?

Use a statistical approach for estimating default

What does a structured-approach model do?

Take an option-pricing approach to estimating default

The best tool to test for nonstationarity is?

Dickey-Fuller Test

What must I remember to do when interpretting a log-linear model model?

Take exponent at the end

How do you test for serial correlation?

Durbin-Watson

When determining whether two series can be used, do we want them to both have, or not have, units roots, and be, or not be, cointegrated?

Either we want neither to have unit roots, or we want both to have unit roots and both be cointegrated. (they can’t have unit roots and not be cointegrated).

What does it mean if two time series has unit roots are are cointegrated?

This means that the error terms from the linear regression are covariance stationary, and linear regression can be used to understand the relationship between them.

What’s the threshold for Durbin-Watson test?

Differs materially from 2 (indicating serial correlation)

What does it mean if the null hypothesis for the Dickey-Fuller test was rejected?

Time series does not have a unit root

What does extraction mean in data wrangling?

Creating a new variable from existing ones

What does veracity refer to?

Credibility and reliability of data sources

What’s the difference between invalid and inaccurate in data preparation?

Invalid data is outside a meaningful range and inaccurate data is not a measure of a true value

How do I determine cheapest-to-deliver obligation?

Must have same seniority as the reference bond, and the lowest trading price