AGEC 622 Forecasting Final Exam

Profit Function

(P*Y)-FC-(VC*Y)

Steps for simulating random variables?

- Be certain that the variable you directly draw is suitable - Estimate parameters - Simulate data under different distributions (norm, uniform, beta, etc.) - Select best distribution based on CDFDEV, Validation tests, etc.

What does USD stand for and how is it used in simulation?

- Uniform standard deviate - Random variable where every interval has an equal probability of being observed (drawn) between 0 and 1 - Used to simulate different types of distributions - =Norm(Mean, StDev, USD)

P-values for betas should be less than ______

0.05

How to make data series stationary?

1. Detrend the data - take the difference b/t consecutive observations 2. Run a Dickey-Fuller Test 3. DF Null Hypothesis states no stationarity - If test yields reject null, data is stationary - Data is stationary when DF test value is < -2.9

t ratios should be greater than _____

1.96

F ratio should be larger than ______ and its P-value less than ______

20.00; 0.05

Bernoulli function

Bernoulli(Prob. of True)

What is First-degree stochastic dominance (FDSD)? Explain how it can (or cannot) be used for ranking risky alternatives.

FDSD occurs when CDFs do not cross each other and have a distribution to the right with higher income at every probability level. With this particular case, all decision makers choose the SAME risky alternative

Suppose you are simulating a revenue (P * Y). What are the consequences if you simulate the price and yield independently and ignore the correlation/dependence between them?

If the correlation between them is negative, mean and variance are biased, and both overstated If the correlation between them is positive, mean and variance are biased, and both understated

Explain the deficiencies of the following risk rankings methods: Means Minimize risk Mean variance Best Case Worst Case

Means - ignores all risk Minimize risk - ignores the levels of returns (mean and relative risk) Mean variance - often difficult to read and often results in tied rankings. Does not work well with non-normal distributions. Best case - ignores overall risk and downside potential risk Worst case - ignores level of returns and upside risk

Z = X*Y If correlation is negative, both mean and variance are biased and you will _________ both mean and risk If correlation is positive, both mean and variance are biased and you will _________ both mean and risk

Overstate; Understate

What is the difference between parametric and non-parametric distributions? Give one example of each.

Parametric distributions have a FIXED form - Shape is dependent on the parameters - Ex. Uniform, normal, beta, bernoulli Non-parametric distributions do NOT have a fixed form dependent on parameters - Ex. Discrete empirical, empirical

Three types of forecasts

Point or deterministic forecast Range forecast Probabilistic forecast

Two types of Models

Structural models - identify the variables (Xs) that explain the variable (Y) we want to forecast, the residuals are the unexplained fluctuations we simulate Univariate (time series) models - forecast using past observations of the same variable

Z = X + Y If the correlation coefficient is positive then the model will _______ the risk for Z If the correlation coefficient is negative then the model will ______ the risk for Z

Understate; Overstate The MEAN is unbiased for X + Y

Deterministic Forecast Formula

Y^i = B0 + B1Xi + B2Zi

Complete probabilistic forecast model

Y^i = B0 + B1Xi + B2Zi + e~i

Linear Trend Model (no risk)

Y^t = alpha^ + B^Tt

Probabilistic Forecast for Y^_T+1

Y~_T+1 = Y^_T+1 + e~

Deterministic forecast of a fixed mean

Yˆ = Y¯ = EYi/N

Polynomial trend model

Yˆt = βˆ0 + βˆ1Tt + βˆ2Tt^2 + βˆ3Tt^3

Beta function

beta.inv(alpha, beta, min, max)

Empirical function

empirical(all residual values)

Stochastic Component

e~i = Yi - Y^i

Normal function

norm(mean, st dev)

Uniform function

uniform(min, max)