1. Meese-Rogoff Redux:
Micro-Based
Exchange-Rate Forecasting
By MARTIN D. D. EVANS AND RICHARD K. LYONS
SHIEH AND HSIEH
International
Economic

2. INTRODUCTION
• Present a micro-based model
• Compare with standard macro model and random walk
• Longer-horizon forecasting
• R.M & K.R.（1983）benchmark
Meese-Rogoff Redux : Micro-Based Exchange-Rate Forecasting

3. INTRODUCTION
• Exchange-rate dynamics from expectational surprises
• Result that the micro-based model better
• Macro model will never explain exchange rate
• Not orthogonal to the evolving real economy
Meese-Rogoff Redux : Micro-Based Exchange-Rate Forecasting

4. MEESE and ROGOFF 1983
S= 𝑎0 + 𝑎1 𝑚 − 𝑚 + 𝑎2 𝑦 − 𝑦 + 𝑎3 𝑟𝑠 − 𝑟𝑠 +
𝑎4 𝜋 𝑐
− 𝜋 𝑐
+ 𝑎5 𝑇𝐵 + 𝑎5 𝑇𝐵 + 𝑢
logarithm of the dollar price of foreign currency
logarithm of the ratio of the U.S. money supply to the foreign money supply
logarithm of the ratio of U.S. to foreign real income
short-term interest rate differential
expected long-run inflation differentia
cumulated U.S. and foreign trade balances
𝑟𝑠
𝜋
𝑦
𝑚
𝑠
𝑇𝐵
Meese-Rogoff Redux : Micro-Based Exchange-Rate Forecasting

5. CONTENTS
1.Forecasting Exchange Rates
2.Forecasting 4 Models
3.Macro
4.Micro
5.Empirical Analysis
6.Conclusions

6. 1.Forecasting Exchange Rates
Why are future changes in exchange rates so hard to forecast ?

7. 𝑆𝑡 = (1 − 𝑏) 𝑖=0
∞
𝑏 𝑖
𝐸𝑡 𝑓𝑡+𝑖 (1)
𝑆𝑡 Log nominal exchange rate
𝑏 Discount factor
𝐸𝑡
𝑓𝑡+𝑖 Current macro fundamentals
Exchange rate
Forecasting Exchange Rates
Meese-Rogoff Redux : Micro-Based Exchange-Rate Forecasting

8. Forecasting Exchange Rates
𝑆𝑡 = (1 − 𝑏) 𝑖=0
∞
𝑏 𝑖
𝐸𝑡 𝑓𝑡+𝑖 (1)
𝑆𝑡 = 𝐸𝑡 𝑓𝑡 +
𝑏
1 − 𝑏
𝐸𝑡∆𝑆𝑡+1
∆S 𝑡+1 =
1−𝑏
𝑏
𝑆𝑡 − 𝐸𝑡 𝑓𝑡 + 𝜀𝑡+1 (2)
𝑆𝑡 = E 𝑡 𝑓𝑡
𝜀𝑡+1 ≡ 1 − 𝑏
𝑖=0
∞
𝑏 𝑖
E 𝑡+1 − 𝐸𝑡 𝑓𝑡+𝑖+1 (3)
Meese-Rogoff Redux : Micro-Based Exchange-Rate Forecasting

9. ∆𝑓𝑡 = ∅∆𝑓𝑡−1 + 𝑢 (4)
𝑆𝑡 − 𝑓𝑡 = ∅ 𝑆𝑡−1 − 𝑓𝑡−1 +
𝑏∅
1 − 𝛽∅
𝑢 𝑡
𝜀𝑡+1 =
1
1 − 𝑏∅
𝑢 𝑡+1
∆𝑆𝑡+1=
1 − 𝑏
𝑏
(𝑆𝑡 − 𝑓𝑡 + 𝜀𝑡+1)
Meese-Rogoff Redux : Micro-Based Exchange-Rate Forecasting
Forecasting Exchange Rates
1 > ∅ > 0 𝑆𝑡 = E 𝑡 𝑓𝑡

10. 2.Forecasting 4 Model
Marco or Micro ?

11. • UIP • Fama
• Micro1 • Micro2
∆𝑺 𝒕+𝟏 = 𝒂 𝟎 + 𝒂 𝒊 𝒕 − 𝒊 𝒕
∗
+ 𝜺 𝒕+𝟏 ∆𝑺 𝒕+𝟏 = 𝒂 𝟎 + 𝒂 𝒊 𝒕 − 𝒊 𝒕
∗
+ 𝜺 𝒕+𝟏
∆𝑆𝑡+1 = 𝜌 + 𝑖 𝑡 − 𝑖 𝑡
∗
+ 𝜀𝑡+1 ∆𝑆𝑡+1 = 𝜂 0 + 1 − 𝜂 𝑖 𝑡 − 𝑖 𝑡
∗
+ 𝜀𝑡+1
∆𝑺 𝒕+𝟏 = 𝒂 𝟎 + 𝒂𝑿 𝒕
𝑨𝑮𝑮
+ 𝒆 𝒕+𝟏
∆𝑺 𝒕+𝟏 = 𝒂 𝟎 +
𝒋=𝟏
𝟔
𝒂𝒋 𝑿𝒋,𝒕
𝑫𝑰𝑺
+ 𝒆 𝒕+𝟏
𝑎0 = 𝜌 , 𝑎 = 1 𝑎0 = 𝜂0 , 𝑎 = 1 − 𝜂
Meese-Rogoff Redux : Micro-Based Exchange-Rate Forecasting

12. 3.Macro Model
The Money-Income And Taylor model

13. 1.the discount factor b is very close to unity
2. Information about future fundamentals arrives simultaneously
to all agents, who in turn revise their forecasts for fundamentals
in unison
Macro Models
Meese-Rogoff Redux : Micro-Based Exchange-Rate Forecasting

14. 𝑓𝑡 = 𝑚 𝑡 − 𝑚 𝑡
∗
− 𝑟 𝑦𝑡 − 𝑦𝑡
∗
+ 𝑞𝑡 − 𝛼𝜌𝑡
𝑓𝑡 ≡ 𝑝𝑡 − 𝑝𝑡
∗
−
1
𝜑0
[𝜌𝑡 + 𝜑1 𝑦𝑡
𝑔
− 𝑦𝑡
∗𝑔
+ 𝜑2(𝜋 𝑡 − 𝜋 𝑡
∗
)]
∆𝑆𝑡+1 → S 𝑡 − 𝑓𝑡
∆𝑆𝑡+1 = 𝑖 𝑡 − 𝑖 𝑡
∗
+ 𝜌𝑡 + 𝜀𝑡+1
Macro Models
Meese-Rogoff Redux : Micro-Based Exchange-Rate Forecasting

15. 𝑖 𝑡 − 𝑖 𝑡
∗
+ 𝜌𝑡 =
1 − 𝑏
𝑏
(𝑆𝑡 − 𝐸𝑡 𝑓𝑡)
∆S 𝑡+1 =
1−𝑏
𝑏
𝑆𝑡 − 𝐸𝑡 𝑓𝑡 + 𝜀𝑡+1 (2)
∆S 𝑡+1 = 𝑖 𝑡 − 𝑖 𝑡
∗
+ 𝜌𝑡 + 𝜀𝑡+1 (5) UIP
Meese-Rogoff Redux : Micro-Based Exchange-Rate Forecasting
Macro Models

16. ∆𝑺 𝒕+𝟏 = 𝒂 𝟎 + 𝒂 𝒊 𝒕 − 𝒊 𝒕
∗
+ 𝜺 𝒕+𝟏
𝑎0 = 𝜌 , 𝑎 = 1
∆𝑆𝑡+1 = 𝜌 + 𝑖 𝑡 − 𝑖 𝑡
∗
+ 𝜀𝑡+1
∆𝑺 𝒕+𝟏 = 𝒂 𝟎 + 𝒂 𝒊 𝒕 − 𝒊 𝒕
∗
+ 𝜺 𝒕+𝟏
𝑎0 = 𝜂0 , 𝑎 = 1 − 𝜂
∆𝑆𝑡+1 = 𝜂 0 + 1 − 𝜂 𝑖 𝑡 − 𝑖 𝑡
∗
+ 𝜀𝑡+1
UIP Fama
Meese-Rogoff Redux : Micro-Based Exchange-Rate Forecasting
Macro Models

17. 4.Micro Model
Aggregate order flow and Disaggregate order flow

18. 𝑆𝑡 = (1 − 𝑏) 𝑖=0
∞
𝑏 𝑖
𝐸𝑡 𝑓𝑡+𝑖 (1)
𝑆𝑡 Log nominal exchange rate
𝑏 Discount factor
𝐸𝑡
𝑚
𝑓𝑡+𝑖 Current macro fundamentals
Expectations conditioned on market-marker’s information at t
Meese-Rogoff Redux : Micro-Based Exchange-Rate Forecasting
A Micro-Based Model
S 𝑡 = (1 − 𝑏) 𝑖=0
∞
𝑏 𝑖
𝐸𝑡
𝑚
𝑓𝑡+𝑖 (7)

19. A Micro-Based Model
S 𝑡 = (1 − 𝑏)
𝑖=0
∞
𝑏 𝑖
𝐸𝑡
𝑚
𝑓𝑡+𝑖
∆S 𝑡+1 =
1 − 𝑏
𝑏
𝑆𝑡 − 𝐸𝑡
𝑚
𝑓𝑡 + 𝜀𝑡+1
𝑚
𝜀𝑡+1
𝑚
≡ (1 − 𝑏)
𝑖=0
∞
𝑏 𝑖
(𝐸𝑡+1
𝑚
− 𝐸𝑡
𝑚
)𝑓𝑡+𝑖+1
Meese-Rogoff Redux : Micro-Based Exchange-Rate Forecasting

20. ∆𝑆𝑡+1 = 𝑎0 + 𝑎𝑋𝑡
𝐴𝐺𝐺
+ 𝑒𝑡+1
∆𝑆𝑡+1 = 𝑎0 +
𝑗=1
6
𝑎𝑗 𝑋𝑗,𝑡
𝐷𝐼𝑆
+ 𝑒𝑡+1
Meese-Rogoff Redux : Micro-Based Exchange-Rate Forecasting
A Micro-Based Model

21. 1. 1.Transactions flows must contain information relevant for
fundamentals.
2. 2. If there is no delay because market-makers can observe
aggregate order flow contemporaneously, then spot rates will be
correlated contemporaneously with order flow
Meese-Rogoff Redux : Micro-Based Exchange-Rate Forecasting
A Micro-Based Model

22. 5.Empirical Analysis

23. Empirical analysis ─ Data
• End-user transaction flows
• Spot rate
• Euro deposit rate
Meese-Rogoff Redux : Micro-Based Exchange-Rate Forecasting

24. Empirical analysis ─Forecast comparisons
• MSE
• Projection stat.
∆ℎ 𝑠 𝑡+ℎ 𝑡 the forecast of ∆ℎ 𝑠𝑡+ℎ ≡ 𝑠𝑡+ℎ - 𝑠𝑡 on day “t”
Step1 recursive h-period out of sample forecasts for non-RW model over the forecasting period
S→ T-h (i.e., ∆𝑠 𝑡+ℎ 𝑡 for S < t ≤ 𝑇 − ℎ)
Step2 regress the forecasts on the realized value
∆ℎ 𝑠 𝑡+ℎ 𝑡 = 𝛽0 + 𝛽 ∆𝑠𝑡+ℎ + 𝑤𝑡+ℎ
Null hypothesis : 𝑠𝑡~RW
Meese-Rogoff Redux : Micro-Based Exchange-Rate Forecasting

25. Meese-Rogoff Redux : Micro-Based Exchange-Rate Forecasting

26. Empirical analysis ─ Result
• MSE ratio is the ratio of mean
squared forecast errors for the
non-RW model to the RW
model.
Meese-Rogoff Redux : Micro-Based Exchange-Rate Forecasting

27. 6.Conclusions

28. Conclusions
Meese-Rogoff Redux : Micro-Based Exchange-Rate Forecasting

29. Conclusions – Future work
• Is the forecasting power here coming from the real economy?
• If the dispersed information framework is the right one,what are the
implications for deep issues?
• what degree is the information being revealed in order flow actually
macroeconomic information?
Meese-Rogoff Redux : Micro-Based Exchange-Rate Forecasting

30. Thank you for your listening!