Economic Analysis Flashcards
Inputs for Mean-Variance Optimization (MVO)
E(r), σ (std deviation), ρ (Correlation)
Framework for determining asset allocation in a portfolio max expected return for a level of risk
Good forecast must be…
unbiased, well researched, with minimum errors and internally consistent (ie. correlations that make sense b/w assumptions)
Challenges in Forecasting
Time Lag Data Issues
1. Official revisions - initial estimates may be revised
2. Definitions & calculation methods change over time (GNP -> GDP)
3, Re-basing index values, cannot mix data w/o adj
Challenges in Forecasting
Data Measurement Error and Biases
4. Transcription errors - problems in gathering & recording data
5. Survivorship bias - data only maintain survivors (upward bias)
6. Appraisal/smoothed data - illiquid markets use appraisal values that do not represent mkt data (smaller correl w/ other assets). Should rescale, increasing dispersion and maintaining mean
Rescale data = multiply σ by the factor given to find new rescaled risk
Challenges in Forecasting
Limitations of Historical Data
7. Regime change - changes in political, regulatory or legal env., creates non-stationarity in data (diff parts have diff underlying assumptions)
8. Data set size - may require long data series, but raises problems (eg. large sets may lead to spurious correl)
9. Ex post (actual) risk can be a biased measure of ex ante (before) risk - evaluate if past asset prices reflect the possibility of a negative event that did not occur (and remove it if it was a mispricing event); only ex ante risk is what is important
Challenges in Forecasting
Biases in Analyst Methods
10. Data mining - actively searching through data to identify a pattern
11. Time-period bias - selecting a specific time frame that explains a patter. Should scrutinize the variable selection process and provide the economic rationale for the variable’s usefulness
Challenges in Forecasting
Failure to “condition information”
12. Failure to “condition information” - If a condition is present in the forecasting scenarion, then you should also condition your variables such as: E(r), σ & ρ in MVO
Challenges in Forecasting
Misinterpretation of Correlations
13. Misinterpretation of Correlations - spurious correlation may exist, or even a third variable may link two variables that have high correlation, but low economic reasoning. Test for partial correlations b/w events using time-series analysis
Challenges in Forecasting
Analyst Behaviour Traps in Forecasting [Solutio]
Challenges in Forecasting
Model Uncertainty
15. Model Uncertainty - Is the model inputs correct? Especially difficult to use models during mkt anomalies. Sampling errors
Statistical Methods of Forecasting
- Historical Return - using historical means as expected return
- Shrinkage Estimators - weighted avg of historical covariation and factor model covariation. Reflects belief and reduces the impact of extreme hist outliers. For return, wgt avg b/w sample avg and indv avg
- Time Series Estimators - using lagged variables to forecast short-term vol
σt²= βσt-1² + (1 - β) ε t²
REMEMBER: Std deviationt = √σt²
ALL SQUARED. The greater the Beta, the greater the remembrance from the last period
Covariance Matrix
σA * σB * Correl(A,B)
σA * σA * Correl(A,A) = σA² * 1
A | B
B
Singer-Terhaar approach to ICAPM
- Calculate expected risk premium for BOTH full integration market and complete segmentation market adding liquidity premium if necessary
ERPi = σ i * Correl(i, GIM) * SharpeGIM = (RPm/σm) + illiquid premium
Remember that Correl(i, GIM) in a completly segmented mkt is equal to 1 (local port is the same of the mkt port)
- Calculate weighted risk premiums for full integration & segmentation
- Calculate exp returns: E(Ri) = Rf + βi [E(Rm) - Rf]
- Calculate each market/asset’s beta: βi = σi * Correl(i, GIM) / σ2GIM
- Calculate covariances: Covi,s = βi βs σ2 and correlations: Correli,s = βi βs (σ2 / σi σs)
Output Gap
Output Gap = trend GDP - actual GDP
- Positive: Recession, no pressure on inflation (BAD THING, due to recessionary characteristic)
- Negative: Expansion, inflationary
Business Cycle
- Recession
Economy, Fiscal and Monetary Policy, Confidence, Capital Markets
- Recession
Economy: Production declines; inflation peaks
Fiscal and Monetary Policy: n.a.
Confidence: Confidence weak
Capital Markets: Short rates declining; bond yields dropping; stocks bottoming and then starting to rise
Business Cycle
- Slowdown
Economy, Fiscal and Monetary Policy, Confidence, Capital Markets
- Slowdown
Economy: Inflation continues to accelerate; inventory correction begins
Fiscal and Monetary Policy: n.a.
Confidence: Confidence drops
Capital Markets: Short-term interest rates peaking; bond yields topping out and starting to decline; stocks declining
Business Cycle
- Late upswing
Economy, Fiscal and Monetary Policy, Confidence, Capital Markets
- Late upswing
Economy: Inflation gradually picks up
Fiscal and Monetary Policy: Policy becomes restrictive
Confidence: Boom mentality
Capital Markets: Short rates rising; bond yields rising; stocks topping out, often volatile
Business Cycle
- Early upswing
Economy, Fiscal and Monetary Policy, Confidence, Capital Markets
- Early upswing
Economy: Healthy economic growth; inflation remains low
Fiscal and Monetary Policy: n.a.
Confidence: Increasing confidence
Capital Markets: Short rates moving up; bond yields stable to up slightly; stock prices trending upward
Business Cycle
- Initial Recovery
Economy, Fiscal and Monetary Policy, Confidence, Capital Markets
- Initial Recovery
Economy: Inflation still declining
Fiscal and Monetary Policy: Stimulatory fiscal policies
Confidence: starts to rebound
Capital Markets: Short rates low or declining; bond yields bottoming; stock prices strongly rising
Inflation at or below exp / Inflation above exp / Deflation
vs.
Cash / Bonds / Equity / Real Estate
Taylor Rule
Used to predict Central Banks moves
Roptimal = Rneutral + ½ (GDPgforecast - GDPgtrend) + ½ (Inflationforecast - Inflationtarget)
If GDP & Inflation are above trend and target, raise interest
If GDP & Inflation are below trend and target, lower interest
Growth trends vs. cycles
Growth trends (long-term) are impacted by population growth, changing demographics, business inventory, gov. structural policies & banking/lending practices. Trends are a key input in DCF.
Cycles (short-term) are alternate sequences of slow and fast growth.
Trends are easier to forecast than cycles, but exogenous shocks can make forecast unpredictable (such as wars, currency, regulation, etc).
Permanent income hypothesis
Consumption is influenced by long-run earnings expectations (somewhat countercyclical even)
GDP trend growth (Cobb Douglas)
GDP trend growth = Δ labor productivity (size and participation) + Δ capital + Δ total factor productivity (TFP)
TFP not observable, residual (ie. you will find it after doing the math)
Emerging Markets - Country Risk Analysis Techniques (for bonds downside risk protection)
- How sound is Fiscal & Monetary policy?
- Economic prospects for the economy
- Is the currency and external accounts under control?
- Is external debt under control?
- Is liquidity plentiful?
- Is the political situation supportive of the required policies?
Leading x Coincident x Lagging Indexes for economy
- *Advantages and Disadvantages of the three approaches to economic forecasting**
1. Econometric Models
- Leading Indicator-Based Approach
- Checklist Approach
Forecasting Currencies
- Purchasing Power Parity (PPP) - Movements in FX is offset by inflation. Currency depreciates w/ higher inflation. [FUNDAMENTAL INDICATOR]
- Relative Strenght (Psych) - Economic growth appreciates FX. Focus on investment flows rather than trade flows. Represents response to news on economy, but nothing about the level of FX. Best used w/ PPP. [TECHNINCAL INDICATOR]
- Capital Flows - Growth and low valuation appreciates FX. This could lead to inflation and at some point, FX depreciation.
- Savings-Investment Imbalances - When savings drop, there is a tendency for a current account deficit which may encourage capital inflow that will depreciate FX
Cobb-Douglas Production Function
Real Output Y = TFP + α * (Capital Growth K) + β * (Labor Growth)
β = 1 - α
L & K are observable. Technical progress or TFP is not and is the residual, that may include innovation, change in tariffs/policies & other.
MUST identify the difference b/w one-time impact vs. structural change for steady-state level forecast.
DDM H-model
High growth that linearly decays to a perpetual growth state
Justified P/E1
Estimated intrinsic value (DDM) divided by year-ahead expected earnings. Used to compare with market P/E
Solve for the perpetual discount rate (r) with Gordon Growth DDM model
r = D1/P0 + g
Problems with Top-Down vs. Bottom-up approach
Bottom-up approach may be too optimistic, being influenced by company mgt
Top-down is based on past trends to forecast data (historical estimates) and the econometric model may be slow capturing inflection points
Relative Valuation Models
Used for tactical allocation decision. All mean-reverting.
1. Fed Model - E1/P S&P vs 10-yr T-note. Best used when prices are at extreme values.
Problems: Ignores equity risk premium, ignores inflation and long-term growth
2. Yardeni Model - E1/P = yB - d x g. d is a weighting factor and yB is the yield on corp bonds.
Problems: Better than yield on gvt bond, but still not the equity risk premium. Assumption used for g (or LTEG) may not be sustainable for perpetuity. d varies over time.
3. CAPE - Cyclically adjusted P/E. Real term Price / 10-yr avg real term earnings. Smooth business cycle, normalizes.
Problems: Change in accounting and low R2
4. Asset-Based Models - Tobin’s q = (MVEquity + MVDebt) / Replacement cost of assets. If q > 1, then the mkt places company`s assets at a higher value than replacement and injecting capital would be profitable. If q < 1, inj K is unprofitable and the company is undervalued
Problems: Change in accounting and low R2
Equity q = MVEquity / Net Worth (NW = replacement cost - liabilities). Different from P/B, since it is not based on historical book value.
Problem: Difficult to get replacement cost estimates, intangible is difficult to value. Low q may persist for long periods of time.
