Forecasting
OGResearch
Tomas Motl
March / ? 2020
2/12
Where are we
• We followed the usual steps in macroeconomic forecasting
1. Pick the right model blueprint for the job
2. Do research about the country, modify the model to fit the country
characteristics
We’ve skipped lots of material here, e.g. parameter calibration
3. Do research on the current situation in the country
4. SKIPPED: Set initial condition for the forecast
5. REMAINING: Use suitable shocks in the model to best represent the
current and expected situation in the country
3/12
Review of model equations
• Standard QPM, equations for the trends are AR(1) processes
• We changed modeling of the FX rate:
− CBA manages the FX rate, pure UIP is not suitable
− We replace the UIP with a ”policy rule”
st = κ1 ∗ (st−1 + ∆star
t − κ2zt
+ (1 − κ1) Et[st+1] + (i∗
t + premt − it)/4 − κ3oilt
∆star
t = ∆zt + πtar
t − π∗
t
− κ1 close to 1
− Note that the ultimate target here is still inflation
• We added influence of oil:
− Oil trend influences the REER trend and the output potential
− Oil gap influences the domestic demand (output gap) and the FX rate
4/12
Review of AZ story
• High dependence on oil for GDP growth, REER
• Policy preference for stable FX rate, but occasional adjustments, not a
strict fixed FX rate
• Interest rates not relevant before 2014
5/12
The 2014-2016 oil shock
• Oil price drop impacted all variables in the economy
• Drop in exports => need to reduce imports and pressure on FX
reserves => real depreciation => nominal depreciation (rather than
decreasing price level)
− Also, depreciations of the trading partners required nominal FX
depreciation (but not real depreciation)
• Imports becoming more expensive relative to domestic production =>
CPI goes up, but less than the FX depreciation
• Lower incomes, banking sector troubles, less employment =>
declining domestic demand
6/12
Initial conditions etc.
• We will not bother with analyzing and tuning initial conditions
− No time, lots of work, ...
− We’ll just trust the model to get it more or less right
• We won’t calculate the necessary REER adjustment, the impact on
GDP, …not difficult, but time consuming
7/12
REER depreciated by about 35%
8/12
External assumptions
• Our model is not supposed to forecast oil price, RUB per USD, …
• We will impose the actual paths (slight cheating) as hard tunes
• External assumptions are important and common source of forecast
error
• QPM model is usually a part of a wider forecasting system
9/12
External assumptions: oil price
10/12
External assumptions: RUB per USD
11/12
Forecasting exercises
• First, we’ll examine the initial conditions
− History (filter) period ends in 2014Q4
− Forecast starts in 2015Q1
• Second, we’ll examine the ”plain forecast”, just external assumptions
without any judgment
• Third, we’ll identify where the forecast is wrong (in our view)
• Fourth, we’ll impose the soft tunes = shocks
− We use a special CSV file for that
• Repeat third and fourth step until we are satisfied
• Important: we need to underestand what the shocks represent!
12/12
Which shocks can we consider?
S
• Strong, permanent depreciation of the REER: shocks to REER trend
• Drop in GDP growth:
− Temporary: shocks to gap
− Permanent: shocks to potential output growth
−
13/12
About the final project
• Two parts: the forecast and the text
• Forecast = forecast report from generated by the codes
• Text = approx 1.5 page explaining the rationale behind the text
− Audience is your macroeconomics teachers: do not know QPM, but
understand jargon and economics
− I’ll provide examples in Study materials; you should not put so many
numbers and so much background, go for something similar but simpler
(e.g. Master’s vs Bachelor’s thesis)
− Describe the whole forecast period - immediate shocks + long term trends
− Test if you really understand what drives the numbers
• To be done in groups again (the previous grouping or a new one)
• Deadline May 20, 8pm
• Ask questions by email / Skype