Title page

Should we use inflationary rate as indicator of currency crisis

 Prologue
The topic of currency crises has been relevant over the last century since they have negative
impacts on economic growth and are very expensive to governments. The crises of the New
Millennium – Mexican Tequila effect (1994), Asian (1997), Russian (1998) and Argentinian
(1999) - stimulated a lot research for creating a signaling mechanism as it affects private
investors and monetary regulators, who are responsible for economic stability. Signaling
mechanisms may benefit both since they can make changes to problematic areas and
therefore escape huge major damages. Early warning systems exist for trying to predict
currency crises before they erupt, which will allow the authorities to act on time. Also, these
models could be used by investors as a part of fundamental analysis and add to estimation of
financial risks.

While the most cited indicators include real exchange rate, exports and imports growth rates,
terms of trade, GDP, M2/international reserves, domestic credit, interest rates, in this paper
we are trying to shed light on the role of inflationary rate in occurrence of currency crisis.
Some literature lists inflationary rate as a significant indicator, while some does not.
Therefore we decided to investigate in what situations inflationary rate could be used and
where it has limited predicting power.

This paper is organised as following. Firstly, we look atexamine the most influential models
that the choice of indicators is based on and we look how inflationary rate (IR) fits into these
models. Secondly, we look at what has been suggested by the literature in terms of prediction
methods and other indicators. Thirdly, we describe the methodology of the signal approach
and we interpret results. Then, we investigate in what episodes IR was useful and analyse the
problems that stop it from being a “good” indicator. Finally, we look at the limitations of this
research and conclude.

Currency crises models

The choice of indicators is based on currency crisis models and it could be noted that
indicators are some way causes of currency crises. Depending on the model, different types
of variables can be chosen. The majority of the research is based on the crises of Latin
America and East Asia. Literature suggests many reasons for currency crisis eruption, but the
most influential models are described below.
To start with, Paul Krugman (1979) developed the “First Generation models” from Stephen
Salant’s and Dale Henderson’s model of speculative attack on gold. The focus of this model
is on fundamental indicators. The model assumes fixed exchange rate and government
intervention when the currency moves away from the set rate. Government may have to
borrow to sustain the international reserves, so it can sell them to revaluate domestic
currency. What the model predicts is that before crisis, there is a gradual decline in
international reserves.
To infer, it can be impossible to peg exchange rate and have expansionary fiscal and
monetary policies. Expansionary fiscal policy can lead to inflation and increase in real
exchange rate, which will require the government to act and sell domestic currency, which
worsens the balance of payments and reduces foreign currency reserves.
Investors are rational economic agents that hold portfolio consisting of domestic and foreign
currencies. They hold the domestic currency as long as they expect it to be feasible.
Therefore, if there is a budget deficit, growing current deficit or high inflation, investors feel
that they might suffer a capital loss due to devaluation and abolition of fixed exchange rate,
they change their expectations and sell their domestic currency. This is described as
“speculative attack”. Speculative attack does not have to be intentionally damaging but it is
when agents sell domestic currency if they feel the weakness in the economy. As a result, this
attack forces government to sell international reserves and until they run out and eventually
the currency is allowed to float freely at a lower rate than fixed rate.
From this model, we can deduce that data related to balance of payments is applicable to
choose as indicators. The advantage of such indicators is that they are quantitative and
published regularly.

Maurice Obstfeld was the first to develop “Second Generation models” (1986) which
explained “self-fulfilling” crises. He focused on expectations of the investors, which could
lead to a crisis even if fundamental indicators and the currency itself were stable. In his view,
speculative attack does not only occur because of insufficient level of reserves to support
deficit on balance of payments but also because speculative attack can be self-fulfilling based
on expectations about government policies. Politics is uncertain and depends on current
economic events, so economic policies respond to the changes in the economy rather than
being set in advance. These policies change expectations, meanwhile policies change in
respond to changed expectations – possibility of multiple equilibria, meaning the economy
can change from one equilibrium to another equilibrium without changes in fundamentals.
Crisis is formed because of the large scale of governmental aims and plans and it has to
choose between keeping the fixed exchange rate and other economic variables. For example,
if unemployment is high, government might want to expand expenditure on unemployment
benefits and since it is treated as monetary expansion and may cause devaluation, it can be
incompatible with the fixed exchange rate policy. Large public and private debts may cause
an attack: investors may become pessimistic if public debt keeps on growing and they may
decide to take their capital out of the country. This is similar with bank runs: if depositors
believe that a bank is going bust, they will clear their accounts. Government cannot refuse to
support the exchange rate as it may lead to political instability, which in turn may affect other
economic variables.
This type of models suggests that fundamental indicators might not be useful at predicting
crises and variables representing government policies or expectations have to be chosen.
However, such indicators are very hard to implement in empirical work since they do not
have quantitative values.

Third generation models show how financial sector and banking sector come with currency
crises. This phenomenon has been known as Twin crises and there is division in opinion on
the order of these two crises. As was listed by Saxena (2004), the work of Stoker (1995)
suggests that currency crisis sets off banking crisis. His research was based on Gold Standard,
i.e. fixed exchange rate, thus if reserves start depleting, it encourages a speculative attack. For
example, in countries with high historical inflation, at the time of speculative attack,
inflationary expectations go up because inflation fighting is not rooted in the system.
Eventually, high interest rates become a consequence after the speculative attack and they
might cause credit crunch and eventually to crisis. Meanwhile, Velasco (1987) argues that
currency crises come after banking crisis. They explain it by that when financial crisis breaks
out, financial institutions have to be bailed out, Central Banks start printing out money, which
leads to depreciation in currency value. To reinforce, Kaminsky and Reinhart (1999) also find
that currency crisis follows from problems in banking sector but they also conclude that
causes of both crises are common: recession that pursues from the boom in economic activity
that was supported by credit, capital inflows and overvalued currency.

Finally, contagious crises have spurred a lot of research. Contagion effect was believed to
have taken place in Asia and Latin America: devaluation of currency in one country caused
devaluation in another. This could have happened because of trade connections, financial
connections herd behavior or confidence crisis. If countries have trade connections,
devaluation in one country increases its competitiveness as exports become cheaper and
imports become more expensive. The demand for trade partner countries’ products will fall
which leads to devaluation in those countries too. For example, speculative attack on Spanish
currency in 1992 devalued it, which made Spanish exports more attractive against Portuguese
exports, which eventually lead to attack on Portuguese currency. Also in 1992, Finland
experienced increase in demand for its exports after devaluation, which negatively affected
Swedish economy since both countries had the same export market. Financial connections
emerge when investors diversify their portfolios, so they invest in different countries. Herd
behavior results in that even if countries do not have financial or trade connections, crash in
stock exchange in one country may cause a crash in another country. This is explained that
investors may want to get rid of weaker currency. A good example of such behavior in stock
exchange crash was in 1987, which is known as Black Monday.

Since we evaluate inflation, we should describe how it fits into these models. In first
generation models, high or increasing levels of inflation can lead to depreciation of currency,
forcing the government to step in, since holding fixed exchange rate. Also, the nature of the
balance of payments crisis is in discrepancy between fundamental monetary politics:
intentions to keep the interest rate and expansionary monetary policy to repay debt, which
leads to higher inflation. In terms of second generation models that were developed by
Obstfeld (1994; 1996), agents form expectations about devaluation and inflation. For the
purpose of this paper, inflationary expectations for the next period equal current inflation plus
some random variable. In twin crises, high volatility of prices reduces the ability of banks to
estimate investment projects and unexpected high inflation can negatively affect bank profits.
Additionally, inflation increases the probability of raising interest rates, which can provoke
capital outflow and lead to credit crunch. Finally, if there is increasing inflation in a foreign
country, demand for currency in that country may be falling, while increasing demand for
domestic currency, which can lead to appreciation and if the country has a fixed exchange
rate, authorities would have to intervene. Similarly, if domestic inflation rises, the probability
of increase in interest rates may cause a capital outflow from the economy. Plus
generallyFurthermore, countries with weak macroeconomic fundamentals, i.e. high inflation,
are typically more inclined susceptive to contagion. However, contagion effects require more
complicated analysis and are not going to be evaluated in this paper, The scope of this study
we will be usingwill take into consideration domestic inflation only.
 Predicting currency crises
Early Warning Systems (EWS) are mathematical models that estimate and predict probability
of crises and are based on the quantity of variables that indicate the environment the current
state of market. Existing literature has proposed several ways to predicting currency crises,
which are discussed below.

Regression analysis
Standard regressions are used for finding the reasons of currency crises. This method does
not allow to find outreveal the start time of crisis but, rather, facilitates the process of finding
it is easier to find out what are the potential countries countries that might experience
speculative attack in case of negative situation in internal and external sectors. This method
assumes that variables have predicting power and each coefficient (β) and the sum of
coefficients are statistically significant from zero. This method follows the formula:
IP = α+Σβ*Ft , where IP is average speed of national currency devaluation and changes in
reserves (gold and foreign currency); Ft is a dummy variable which takes on value 1 if
variable crosses a certain threshold or not, i.e. issues a signal. Thresholds are chosen by the
authors, and are thus very subjective.
Some authors also use logit/probit analysis that allows predicting for the prediction of the
probability P(i,t) of a crisis taking place at a certain time t in a country I and is a function of a
list of indicators. P(i,t) takes on value 1 when crisis happened and 0 otherwise. The name
comes from what function authors choose, the model becomes probit or logit. This method
allows interpreting result as probability of a crisis since value will always be between 0 and 1
and it also allows to usethe usage of all variables available.
The disadvantage of such method is that the method procedure is complicated and, like with
regression analysis, many variables have to be accounted in order to eliminate correlation
between variables.
This method was popularized by Eichengreen, Rose and Wyplosz (ERW) (1995). In their
paper they followed probit regressions technique and used quarterly data for countries that
were under Exchange Rate Mechanism and non-members in order to compare
macroeconomic variables during vulnerable episodes during the “tranquil” episodes. They
hoped that the behavior of these variables would be different between the two periods
because of the inconsistent macroeconomic policies implemented by the government. To
their disappointment, there was no difference in behavior and thus they hypothesized that
currency crises react to changes in expectations. Furthermore, they found that money growth
and inflation do not follow first generation crisis models prediction.
Futhermore, Moreno (1995) had a sample of the Pacific Basin economies and he found that
exchange rate market vulnerability is linked large budget deficits and rapid rates of growth in
domestic credit. He added that there are episodes when relatively high inflation and slow
growth made it hard for the government to keep a stable exchange rate.
Finally, Further, Frankel and Rose (1996) followed probit regression technique and used
annual data of developing countries. Surprisingly, they found that first-generation variables
such as current account and government budget deficit do not have any predicting power for
currency crashes. They concluded that these crashes are associated with high foreign interest
rates, low output growth, high domestic credit growth and low ratio of foreign direct
investment/debt.

The Signal approach

Kaminsky, Lizondo and Reinhart (1998) (KLR) proposed a monitoring method which looked
at single indicators. The method is based on the idea that some variables may act unusually
prior to a crisis. An indicator should pass a certain threshold, which is chosen by an author
specifically and is specific to the sample, in order to become a signal. Choosing an
appropriate threshold is critical for this method to reduce false alarms without undermining
the warning function. Thus if we examine these variables we will be able to predict
vulnerability of the economy. We choose this method to investigate the effectiveness of
inflation rate as an indicator of currency crisis for this dissertation for several reasons. Firstly,
it is more general non-parametric approach to predicting currency crises since it does not
assume a specific model of a currency crisis – as can be seen from the discussion of currency
crises models, inflation rate is not a significant factor in any model, so a method that does not
assume any models is suitable. Secondly, predicting power of any indicator using this method
is analysedanalyzed on individual level and we are free from any correlation issues that could
arises in regression analysis.

Indicators considered by literature

Since indicators are primarily derived from currency crisis models, we can draw 4 main
groups of indicators: macroeconomic or fundamental; linkage indicators that show how
different sectors and economies are connected with each other; and indicators show the mood
in the markets.
Some works statistically analyse macroeconomic indicators, for example, Berg and Patillo
(2005). This method has a number of drawbacks. As one of their indicators they use changes
in stock prices. National currency devaluation expectations will lead to the increase in interest
rate differential for assets that are kept either in foreign or national currencies. If economic
downturn is expected, investors will want to take away their loans and debt instruments,
which will eventually be reflected on stock prices. Indicators of vulnerability as addition to
macroeconomic indicators, reflect how protected a currency is against speculative attacks.
For example, the ratio of short-term external/internal debt to the level of foreign currency
reserves could serve as a good indicator of vulnerability.

The other “contagion” approach to indicators is how tied are economies together through, for
example, financial (Foreign Investment and common lenders) and trade linkages. The
probability of crisis can increase if there is a crisis in the region, despite having strong
fundamentals, which is dictated by the contagion channels. Financial linkages can be
explained by investors’ diversification portfolio. In the globalisation context, vulnerability in
one country can lead investors to reconsider their risks and thus alter their portfolios. On a
national level, contagion can be spread through inter-bank operations. For example, if central
banks are slow to refinance banking sector and if there is a high demand for liquidity from a
group of banks, it is possible that they will not be able to meet the requirement deadline of
other banks, which will worsen the financial state of the banks, which had no liquidity
problem before.

Market expectations indicators are a separate branch of research. Market expectations are
very significant for forming exchange rate. However, they are very difficult to measure and
also, they indicate crisis in later stages rather than in early stages. For example, herding
behaviour played an important role in spreading the crisis in Asia in 1997.
The vailable Lliterature also recognises institutional factors such elections, war, accounting
standards but they are hard to measure and for that reason are not usually used as indicators.

Data, methodology, results

Data
The data was taken from the International Financial Statistics website. The set is monthly
from 1970 to 2007 for the majority of countries, but some countries, for example, post-USSR
countries have shorter time-horizon. As frequency of crises in one country is low, our sample
had to be expanded to more than 1 country. The selection criterion for countries was at least 1
crisis episode in the past and we limited our sample to an arbitrary number of countries from
different regions.
Variables Source: International Financial Statistics
Foreign reserves Total reserves minus gold, units: millions of US dollar
IFS line 1L.d
Nominal Exchange rate End of period rate, units: national currency per US dollar
IFS Line 00ae.
Inflation rate CPI %change from a year before
IFS Line 66xzf

Methodology
Defining crisis
First of all, a definition of crisis should be given. There are several definitions of currency
crises in empirical literature. Eichengreen, Rose and Wyplosz (ERW, 1999) define it in their
1994-1996 paper as large movements in exchange rates, interest rates and international
reserves. Frankel and Rose (FR, 1996) state that crisis is happening if currency depreciates by
at least 25% in the nominal exchange rate that also exceeds the previous year's depreciation
level by at least 10%. Sachs, Tornell and Velasco ( define crisis as the weighted average of
the percent depreciation of the nominal exchange rate and the percent decrease in reserves.
For investigation, a more statistical definition is needed.

KLR used the definition used by ERW and we are going to follow this definition but with
slight changes like in work of Hali J. Edison (2000), as we believe are relevant and are more
beneficial to the analysis. We define currency crises as a deviation by 2.75 standard
deviations, δ, from index pressure in the foreign exchange market IPt at month t. This
measure is known as Exchange Market Pressure (EMP).

EMP= β1%Δet – β2%Δrt

%Δet – monthly percentage change in foreign currency per US dollar at time t; %Δrt –
monthly percentage change in foreign reserves; β1 and β2 are weights which are chosen that
way that will not allow any variable to dominate the index since their volatilities are different.
Literature largely suggests that this can be achieved if β1 is 1 and β2 is the ratio of the standard
deviation of the rate of change of the exchange rate to the standard deviation of the rate of
change of reserves (δe/δr).
Crisis is then defined as:
Crisis = 1 if EMPt>3 δemp + meanemp
Crisis = 0 otherwise
Where δemp and meanemp are derived from the sample and 3 δemp + meanemp is a threshold.
Coefficient of 2.75 is arbitrary and chosen by consideration of authors and seeks to optimize
the number of crisis episodes. Higher the coefficient, more extreme crisis episode will only
be considered, while lower coefficient will indicate “vulnerability” episodes, when an
economy experience the increased likelihood of currency attack but no crisis is necessary. In
literature, many coefficients have been used – 1,5; 2; 2,5 and 3. After trying different
coefficients, we decided to use 3 and we saw it fir best for our data and allows us to capture
more significant crisis episodes rather than vulnerability episodes. Notably, in order not to
count the same crisis episode twice, we excluded observations where two crises occur in 6
successive months (2 successive quarters). Thus there are two types of periods: “crisis” and
“tranquil”, which we will be using in assessing indicator.
Countries like, for example, Argentina, which experienced hyperinflations, were treated as is
proposed in KLR method. The data was divided into two sub-samples, which would have
different means and standard deviations: “tolerable” inflation and “hyper” inflation periods if
inflation rate was exceeding 150% 6 months before. In light of the topic of this dissertation,
this might seem implausible as hyperinflation can be considered as a currency crisis because
of the accelerating devaluation/depreciation in value. However, we are interested in
speculative attack episodes and hyperinflation as a phenomenon on its own is not considered
to be a speculative attack. Thus, we have to adjust the index to get more correct results for
crisis episodes.

Thresholds for indicators

After defining crisis episodes, IR had to be evaluated and we used the scheme described
further.
An indicator is said to issue a signal when it passes a threshold, i.e. deviates from “normal”
trend. An optimal threshold needs to be chosen using specific criteria in order to minimize
false alarms but not disadvantage true signals. Following KLR (1997) and Edison (2003), we
define threshold relatively to the 10th percentile of the distribution of the indicator”, meaning
that the 10th percentile will be used as a reference point for obtaining a threshold.
To optimize the thresholds, KLR propose to use the ratio of good-to-bad signals. For this
purpose, we need to divide indicators on 4 groups:
 Crisis No crisis
(24 months) (24 months)
Signal was issued A B
No signal was issued C D

Ideal indicators would be A and D. Good signals are those that cross the threshold before 24
months of a crisis breakout. Thus, 24 months is called “crisis window”. Later we will look at
additional “crisis windows” – 18, 12, 6 and 3 months to see if signals persist and how
indicators evolve at upcoming crises. Using the table above, following ratios will help us
analyse indicators:
A/(A+C) – measures how effective is a variable at producing good signals. It expresses the
number of good signals as the percentage of the number of months in which good signals
could be produced. For example, in 24 months before crisis, a variable passed the threshold 3
times, the ratio will be 3/24. The higher the ratio, the better is indicator.
B/(B+D) – measures the tendency of an indicator to issue bad signals. It shows the
percentage of bad signals to the number of months when the indicator could have been
issued. Similarly to the previous ratio, of a variable passed the threshold 3 times but there was
no crisis in the next 24 months until the “crisis prediction” period, the ratio is 3/24. The lower
is the ratio, the better is the indicator.
[B/(B+D)]/[A/(A+C)] – noise-to-good signals ratio (“noisiness”). This ratio combines both
previous ratios and shows how an indicator can produce good signals while avoiding the bad
ones. The lower the ratio is, the better an indicator is. The last ratio is the most significant at
assessing indicators.
Additionally, we measure the conditional probability of a crisis - P(signal/crisis) - if signal
was issued with A/(A+B) and unconditional probability - P(crisis) - (A+C)/(A+B+C+D). This
measure will help to analyse if the indicator is valuable at predicting crisis. For it to provided
valuable information, P(signal/crisis) has to be higher than P(crisis) as it shows

Results
Crisis episodes
Bold – crisis episode, while “normal” is when there are less than 6 months apart.
Country EMP Speculative attack episode
threshold
Argentina
01/1970-12/2007
01/1970-12/1990 162.3809268 1975m06 165.9836486
1982m07 135.7565701
1989m04 419.489055
1989m12 211.4224201

01/1991- 12/2007 44.22034 1991m03 52.47994

2002m01 49.15863
2006m01 52.84164
Bolivia 163.3782767 1981m06 237.7831
01/1970-12/2007 1982m04 207.4644
574.7902 1985m09 1381.178
1988m01 307.7962
Brazil 39.88086089 1982m09 73.12342349
01/1970-12/2007 1983m01 58.34202051
1999m01 52.09227585
Chile 39.32073685 1971m07 208.3170331
01/1970-12/2007 1972m09 69.75209202

165.0745238 1973m05 165.492853

Colombia 9.337679893 1985m01 9.36704982
01/1970-12/2007 1995m08 9.560834876
2002m07 9.862211857
2007m08 10.33962156
Denmark 12.83408842 1980m03 12.88558487
01/1970-12/2007
Ecuador 57.186831 1983m10 59.76919867
01/1970-04/2000 1972m05 62.87371039
Finland 12.68186579 1991m03 13.32416814
01.1970-12/1998 1992m09 15.8725279
India 6,658088739 1991m07 21,52919024
01/1970-12/2007 1995m09 6,815326633

Indonesia 34.34695754 1978m11 54.24878774

01/1971-12/2007 1986m09 49.24181583
1998m05 34.94297682
1998m06 42.41982536
Jordan 6.252296674 1988m06 10.94648538
01/1970-12/1998 1988m10 17.4700224
1989m02 8.652153338
Korea 14.46033011 1980m01 22.30290033
01/1970-12/2007 1997m11 29.54000356
1997m12 52.97668663
Mexico 32.72225178 1976m09 73.85512
01/1970-12/2007 1976m10 43.43929
1982m02 77.12613
1982m12 91.97694057
1994m12 51.80151367
Malaysia 9,361865235 1998m01 19,64166326
01/1970-12/2007
Paraguay 24.97281916 1984m06 50.1936908
01/1970-12/2007 1985m03 44.8936208
1986m12 59.82003993
1989m03 66.29649871

Peru 117.2914102 1976m06 119.9559536

01/1970-12/2007
Philippines 15.92038949 1970m02 38.31627419
01/1970-12/2007 1983m10 30.1427588
1984m06 35.19688596
1986m02 18.00329791
1997m12 18.65252457
Russia 28.04681777 1998m09 103.2194813
12/1994-12/2007
Singapore 7.77598544 1997m12 8.39407597
01/1970-12/2007 1998m01 7.937609139
Sri Lanka 19,82434183 1998m07 86,32502831
01/1970-12/2007
Sweden 12.9014727 1977m08 13.28981196
01/1970-12/2007 1992m11 27.95989384
1993m02 13.2884939
Thailand 10.23240133 1997m07 27.57473644
01/1970-12/2007 1997m08 15.09696024
 1997m12 15.80722208
Turkey 24,80825208 1994m04 53,97611892
01/1970-12/2007 2001m02 36,10834997

United Kingdom 13.15024481 1992m09 13.82282719

01.1970-12/2007 1992m10 15.21458229
Uruguay 31.21760128 1982m11 57.3590669
01.1970-12/2007 1982m12 67.35393342
2002m07 51.84358792
Venezuela 34,327373 1989m02 154,4827586
01/1970-12/2007 1996m04 60,79310345

2002m02 38,78351864
Authors calculations.
Total crisis episodes (highlighted in bold): 61.

Inflationary rate analysis - statistics

Number of Crises A: B: C: D:
crises Predicted Signal Signal No No signal
+crisis +no signal +no crisis
crisis +crisis
Argentina 7 2 11 0 130 308
Bolivia 4 2 18 21 39 350
Brazil 2 2 3 20 45 386
Chile 3 2 7 45 31 370
Colombia 4 0 0 16 96 340
India 2 1 2 35 22 395
Indonesia 3 1 3 32 69 349
Jordan 2 1 1 14 31 324
Malaysia 1 0 0 26 23 406
Mexico 4 1 3 34 55 350
Peru 1 0 0 21 24 410
Philippines 4 2 19 17 57 359
Sri Lanka 1 1 3 43 21 388
Thailand 1 0 0 7 24 424
Turkey 2 0 0 27 24 403
Uruguay 2 0 0 2 48 404
Venezuela 3 2 20 23 52 358
United
Kingdom 1 0 0 3 24 428
Sweden 2 2 6 29 43 376
Russia 1 0 0 3 24 152
Denmark 1 1 1 21 23 410
Ecuador 2 1 2 26 46 380
Finland 2 0 0 43 48 363
Korea 2 0 0 4 50 388
Paraguay 4 3 10 20 73 339
Total 61 24 109 532 1122 9160

Inflationary rate – statistics continued

a/(a+c) b/(b+d) P(crisis/signal) “Noisiness” Probability
0,07801
Argentina 4 0 1 0 0,31403118
0,31578 0,05660 0,17924528
Bolivia 9 4 0,461538462 3 0,13317757
0,04926
Brazil 0,0625 1 0,130434783 0,78817734 0,105726872
0,18421 0,10843 0,58864027
Chile 1 4 0,134615385 5 0,08388521
0,04494
Colombia 0 4 0 #DIV/0! 0,212389381
0,08333 0,08139 0,97674418
India 3 5 0,054054054 6 0,052863436
0,04166 2,01574803
Indonesia 7 0,08399 0,085714286 1 0,158940397
1,32544378
Jordan 0,03125 0,04142 0,066666667 7 0,086486486
0,06018
Malaysia 0 5 0 #DIV/0! 0,050549451
0,05172 0,08854 1,71180555
Mexico 4 2 0,081081081 6 0,131221719
0,04872
Peru 0 4 0 #DIV/0! 0,052747253
0,04521 0,18085106
Philippines 0,25 3 0,527777778 4 0,168141593
0,09976 0,79814385
Sri Lanka 0,125 8 0,065217391 2 0,052747253
0,01624
Thailand 0 1 0 #DIV/0! 0,052747253
0,06279
Turkey 0 1 0 #DIV/0! 0,052863436
0,00492
Uruguay 0 6 0 #DIV/0! 0,105726872
0,27777 0,06036 0,21732283
Venezuela 8 7 0,465116279 5 0,158940397
United 0,00696
Kingdom 0 1 0 #DIV/0! 0,052747253
0,12244 0,07160 0,58477366
Sweden 9 5 0,171428571 3 0,107929515
0,01935
Russia 0 5 0 #DIV/0! 0,134078212
 0,04166 0,04872
Denmark 7 4 0,045454545 1,16937355 0,052747253
0,04166 0,06403 1,53694581
Ecuador 7 9 0,071428571 3 0,105726872
0,10591
Finland 0 1 0 #DIV/0! 0,105726872
0,01020
Korea 0 4 0 #DIV/0! 0,113122172
0,12048 0,46239554
Paraguay 2 0,05571 0,333333333 3 0,187782805
0,08854 0,05489 0,61991162
Total 6 1 0,170046802 7 0,105726872
“#DIV/0!” means that no good signal was issued at all

Comparison with exports growth rate

As we thought that results may not be sufficient enough on its own, we decided to compare
them with another indicator to give us a reference point what ratios can be. Assuming that
literature is correct, we chose exports growth rate as it has been identified by many authors as
a leading indicator and additionally it is an easily obtainable statistic. In original KLR
conclusion, exports called 85% crises and had noisiness ratio of 0.42. We took the dataset
from IFS website, line 70.dzf.

Exports growth rate

Number of Crises A: B: C: D:
crises predicted Signal Signal No signal No signal
+crisis +no crisis +crisis +no crisis

Argentina 7 6 18 18 125 288

Bolivia 4 4 8 20 73 249
Brazil 2 2 3 27 45 378
Chile 3 3 8 23 31 390
Colombia 4 2 5 32 91 323
India 2 2 3 25 45 381
Indonesia 3 2 7 12 66 368
Jordan 2 2 6 31 23 386
Malaysia 1 1 3 2 21 429
Mexico 4 3 3 2 78 350
Peru 1 1 1 4 23 410
Philippines 4 3 2 11 72 367
Sri Lanka 1 1 3 26 21 405
Thailand 1 0 2 37 24 430
Turkey 2 1 3 6 45 400
Uruguay 2 1 2 12 46 393
Venezuela 3 3 4 23 69 357
United
Kingdom 1 1 1 35 24 430
Sweden 2 2 3 7 45 398
Russia 1 1 2 8 22 158
Denmark 1 1 2 15 22 414
Ecuador 2 2 11 20 37 384
Finland 2 2 3 31 38 381
Korea 2 1 2 4 46 401
Paraguay 4 4 10 14 265 368
Total 61 51 115 445 1397 9238

Exports growth rate - continued

a/(a+c) b/(b+d) P(crisis/signal) "Noisiness" Probability
0.12587 0.05882 0.46732026
Argentina 4 4 0.5 1 0.318485523
0.09876 0.07434 3.32098765
Bolivia 5 9 0.285714286 4 0.231428571
0.06666
Brazil 0.0625 7 0.1 8.4375 0.105960265
0.20512 10.5897435
Chile 8 0.05569 0.258064516 9 0.086283186
0.05208 0.09014 3.69791666
Colombia 3 1 0.135135135 7 0.21286031
0.08333 0.03496
Denmark 3 5 0.117647059 17.875 0.052980132
0.22916 0.04950 8.41666666
Ecuador 7 5 0.35483871 7 0.10619469
0.07317 0.07524 10.0487804
Finland 1 3 0.088235294 9 0.090507726
0.03157 5.20547945
Indonesia 0.09589 9 0.368421053 2 0.161147903
0.20689 0.07434 14.3793103
Jordan 7 1 0.162162162 4 0.065022422
0.04166 0.00987
Korea 7 7 0.333333333 8.4375 0.105960265
0.03703 0.00540 4.56790123
Mexico 7 5 0.6 5 0.179600887
0.03636 0.03664 1.38909090
Paraguay 4 9 0.416666667 9 0.418569254
Peru 0.04166 0.00930 0.2 17.9166666 0.052863436
 7 2 7
0.02702 0.02925 5.08108108
Philippines 7 5 0.153846154 1 0.164444444
0.08333 0.04819 6.91666666
Russia 3 3 0.2 7 0.126315789
0.04166
Singapore 7 0 1 #DIV/0! 0.052863436
0.01728
Sweden 0.0625 4 0.3 8.4375 0.105960265
0.07692 0.07922 17.9615384
Thailand 3 9 0.051282051 6 0.052738337
United 0.07526
Kingdom 0.04 9 0.027777778 18.6 0.051020408
0.04166
Uruguay 7 0.02963 0.142857143 8.4375 0.105960265
0.07581 0.04486 0.59176492
Total 5 5 0.215982721 4 0.140170032

When counting the signals, what was noticeable about the inflationary signal is its clustered
distribution – if it passes a threshold this month, it is very likely that the next month it will
pass it too. Thus signals, false or good, come in groups and that is why the number of noise
can be very large. Nevertheless, on all accounts we see that inflation rate is less efficient than
exports. To start with, exports called 51 crises 61, which is 83%, while inflation called only
24, making it 39.3%. By “calling” we mean that an indicator issued a signal at least once
before crisis. “Noisiness” ratio is higher for inflationary rate, pointing to the larger number of
“false” alarms produced per 1 “good” alarm. The difference, however, is relatively small,
about 3%, but this is counteracted by the noisiness ratio and both probabilities.

Different types of inflationary behaviour before crisis

While this shows that IR is a less competent indicator than exports, it does not prove that it
should not be used at all. For this purpose, we decided to investigate the behaviour of
inflation around some crisis episodes more closely. We chose speculative attack summaries at
random.

Sweden
At first, we will consider Sweden that experienced falling inflationary rate before crisis.
 1990m1 1991m0 1991m0 1991m0 1991m0 1991m0 1991m0
2 1 2 3 4 5 6 1991m07
10.944 10 12.5563 9.93184 10.6725 10.126 10.0873 9.07781
1991m0 1991m0 1991m1 1991m1 1991m1 1992m0 1992m0
8 9 0 1 2 1 2 1992m03
8.1584 8.11321 7.82568 7.94021 7.90089 5.16217 2.35556 2.43578
1992m0 1992m0 1992m0 1992m0 1992m0 1992m0 1992m1
4 5 6 7 8 9 0 1992m11
2.11361 2.06775 1.98238 1.80537 2.02911 2.35602 2.17297 1.25487
Source: IFS
During that period, Sweden had fixed exchange rate. Sweden was experiencing low GDP
growth since 1970. Deregulation of credit markets in 1985 put an end to expansionary
monetary policy and led to increasing debt. In 5 years, debt-to-GDP increased from 85% to
135%, while housing prices and stock exchange index were rising too. As a result Swedish
economy became vulnerable to shocks. Competitiveness was threatened by high IR at the end
on 1980s, which led to overvaluation of currency and eventually to fall in exports. Authorities
had to raise nominal interest rates to offset the inflation.
In the second half of 1980s, the price of accumulated assets rose by more than 125%, which
helped to form speculative bubble. Private savings dropped by 7% and even reached negative
values. Inflation was rising and economy started to heat up. Current account deficit was
supported by capital outflow, causing the private sector short-term debt increase, which was
kept in foreign currency.
Reformed tax system resulted in increased interest rates after paying taxes. At that point,
assets started to lose in value and economic activity fell, for example, from 1990 to 1993
GDP of Sweden dropped by 6%. Unemployment rose to 12% and government debt was 12%
of GDP. All these factors were signaling of the coming banking and currency crises.
We can see that in February 1991, inflation signaled of the coming crisis
The role of inflation can be expressed as since it was accelerating, it forced the authorities to
produce policies that would deprive the economy and increase the probability of crisis.
Mexico
On the other hand, Mexico experienced different type of inflationary behaviour before the
crisis in 1982.
1979m0 1979m0 1979m1 1979m1 1979m1 1980m0 1980m0
8 9 0 1 2 1 2 1980m03
17.8606 17.9437 18.6002 18.8805 20.0376 21.6364 22.6703 23.519
1980m0 1980m0 1980m0 1980m0 1980m0 1980m0 1980m1
4 5 6 7 8 9 0 1980m11
24.4308 24.8055 26.0051 27.8716 28.7261 28.5362 28.2134 28.8109
 1980m1 1981m0 1981m0 1981m0 1981m0 1981m0 1981m0
2 1 2 3 4 5 6 1981m07
29.7806 27.8027 27.9766 28.0601 28.7122 28.6011 27.8344 26.5522
1981m0 1981m0 1981m1 1981m1 1981m1 1982m0 1982m0
8 9 0 1 2 1 2 1982m03
26.5201 27.4472 28.3733 28.5626 28.6836 30.8187 32.7055 34.7121
1982m0 1982m0 1982m0 1982m0 1982m0 1982m0 1982m1
4 5 6 7 8 9 0 1982m11
38.9284 44.5342 49.3893 54.3841 68.2004 73.9458 79.0275 84.5301
Source: IFS
Economic situation from 1977 to 1981 was supported by government spending, which was
not assisted with large amounts of income. There was a significant budget deficit and current
account deficit. Trade liberalisation and real exchange rate growth reduce real cost of imports
and were enforcing deficit current account deficit. In 1981, quotas on many imported goods
were introduced; however, imports still grew at 15.2%. While oil exports were bringing
dollar income into the economy, while increased real exchange rate negatively affected all
other export goods. Rise in current account deficit was sponsor was subsidized by rising
external debt. Repayment of debt, including short-term debt, was 80% of capital inflow. In
1982, despite government announcement of 40% devaluation, monetary and fiscal expansion
continued and government debt rose to 17.6% of GDP. Economy was experiencing capital
outflow.
IR in this case was accelerating because of the expansionary policies that were adopted by the
authorities. But it also did not signal the crisis 24 months before. Larger extract with more
than 24 months was given to get a bigger picture how inflation was increasing.
Turkey
The Turkish crisis of 1994 originated in 1989, when capital liberalization was said to be
finished. This liberalization was assisted with fiscal deficit. Government debt was rising from
1988 until 1993 because it increased its spending on agricultural subsidies, military, state
institutions, while the revenue was not increasing.
1993m0 1993m1 1993m1 1993m1 1994m0 1994m0 1994m0 1994m0
9 0 1 2 1 2 3 4
68.2452 67.2003 69.6184 71.076 69.6466 72.9641 73.6455 107.446
1993m0 1993m0 1993m0 1993m0 1993m0 1993m0 1993m0 1993m0
1 2 3 4 5 6 7 8
59.7688 58.159 58.0102 58.9701 65.0286 67.2357 73.1157 71.2437
1992m0 1992m0 1992m0 1992m0 1992m0 1992m1 1992m1 1992m1
5 6 7 8 9 0 1 2
69.862 65.7916 65.803 65.5029 67.6658 69.184 68.5932 65.9666
Source: IFS
In 1991, the government announced anti-inflationary actions through decreased public
expenditure but it was counteracted with high interest rates as a saving aid. In 1992, the
government used all possible finances from the Central Bank and eventually had to borrow in
foreign markets in order to finance the debt repayments. According to Ozatay (1996), the
economy was already vulnerable in 1992, the crisis was only postponed by the intervention in
domestic borrowing market. He then concludes that this crisis happened because of debt
mismanagement.
Analogically to Mexican crisis, this crisis fits the description of the first generation model,
since there were deteriorating fiscal factors and at the end, domestic credit expansion lead to
currency depreciation despite it not being fixed.

Alternatively, Turkey had another experience too in 2001. After the fall in GDP in 1999,
Turkey was experiencing economic boom and GDP grew to 7.4% in 2000, but then slumped
again in 2001, while current account deficit was widening at increasing rate. The economy
was also experiencing high inflation, which forced the authorities to adopt dis-inflation plan
in 2000. Erinç Yeldan (year unknown) argues that this plan caused the Turkish Lira
depreciation in 2001. The programme limited the actions of the Central Bank so it could only
operate expansionary monetary policy to increase the stock of net foreign assets. Since
Turkey had floating exchange rate, this meant that domestic interest rate was dependent on
foreign capital and that it was more vulnerable to speculation.
2000m0 2000m0 2000m0 2000m1 2000m1 2000m1 2001m0 2001m0
7 8 9 0 1 2 1 2
56.214 53.1652 48.9583 44.4365 43.7573 39.0336 35.9175 33.4219
1999m1 1999m1 2000m0 2000m0 2000m0 2000m0 2000m0 2000m0
1 2 1 2 3 4 5 6
64.5518 68.7921 68.8783 69.7465 67.8962 63.8248 62.6747 58.6182
1999m0 1999m0 1999m0 1999m0 1999m0 1999m0 1999m0 1999m1
3 4 5 6 7 8 9 0
63.5374 63.855 62.9667 64.2696 65.0035 65.4046 64.27 64.6954
Source: IFS
We can see that authorities while trying to reduce IR, undermine the stability of exchange
rate by adopting special plan. For that reason, Turkey saw the same picture as Sweden but
with the floating exchange rate.

India
Current account deficits were accumulating from 1980s, while reserves were depleting. Two
shocks negatively affected Indian economy, further forcing it into debt and thus making the
economy more vulnerable. First, it was the increase in oil prices in 1990 but then exports
growth slowed down because of the slow economic growth of the trading partners.

1990m1 1991m0 1991m0 1991m0 1991m0 1991m0 1991m0 1991m0

2 1 2 3 4 5 6 7
13.7143 16.092 15.4286 13.5593 12.2222 12.0879 12.973 13.2275
1990m0 1990m0 1990m0 1990m0 1990m0 1990m0 1990m1 1990m1
4 5 6 7 8 9 0 1
7.78443 7.69231 8.82353 9.88372 9.1954 8.52273 10.7955 12.5
1989m0 1989m0 1989m1 1989m1 1989m1 1990m0 1990m0 1990m0
8 9 0 1 2 1 2 3
6.77271 7.19601 4.98176 4.34781 5.02332 5.45455 6.06061 6.62651
Source: IFS
Cerra, Valerie and Sweta Saxena (2002) argue that while Indian crisis in 1991 is not
explained well by first generation model, since India imposed capital controls, it shares some
characteristics with it: expansionary fiscal policy do not fit fixed exchange rate regime.
Therefore, inflationary rate was increasing.

Russia
After the USSR collapsed, Russia pegged ruble against dollar in order to control inflation.
However, Russian government faced three problems. First, until 1995 the internal
government debt was rising through deposits. Government had options: default on internal
debt, print out money with the risk of hyperinflation or opening up the market for foreign
investors (the chosen route). There was high expenditure – 60% of tax income. Second, there
was inefficient monetary policy – inflation was higher than money supply, which meant real
depreciation, while there were high interest rates. Third, there was overvaluation of ruble,
which meant lower competitiveness.
1998m0 1998m0 1998m0 1998m0 1998m0 1998m0 1998m0 1998m0
2 3 4 5 6 7 8 9
9.4132 8.55937 7.9403 7.45121 6.43651 5.62319 9.63436 52.1537
1997m0 1997m0 1997m0 1997m0 1997m1 1997m1 1997m1 1998m0
6 7 8 9 0 1 2 1
14.4675 14.6941 14.8667 14.1528 13.0007 11.5335 11.0546 10.1179
1996m1 1996m1 1996m1 1997m0 1997m0 1997m0 1997m0 1997m0
0 1 2 1 2 3 4 5
27.1594 23.9718 21.8154 19.777 18.2936 16.6667 15.3581 14.5664
Source: IFS
As can be seen, IR was dropping before the speculative attack. Russian can be characterised
by an overlap between balance of payments crisis since government was running deficits,
depleting reserves and printing out money, while it also had high interest rates to try to curb
inflation.

From investigating the examples above and inflationary rates in countries from our sample,
we found that they behave in a variety of ways – they can either accelerate or fall when an
economy is approaching the crisis date. Sometimes, acceleration can grow into hyperinflation
– for example, Argentina, Brazil, Chile, Kazakhstan and Ukraine. Summaries above give an
illustration to what is happening in the economy. 17 out of 44 studied crises experienced
falling inflationary rate beforehand. Interestingly, when an economy is following a second
generation currency crisis scenario, it may seem that inflationary rate is falling. Although it
does so for Sweden in 1991 and Turkey in 2001, there is no empirical proof that it is always
the case.
Inflation increases not because of crisis – reserves are low and inflation is lower (time-lag)
However, one of the reasons for such inflationary behaviour, as proposed by research of Stein
and Streb (1993) on countries with high historical inflation, is that governments trade
between current and future inflation in proximity of elections and they use reserves for that
purpose. This theory is consistent with the “self-fulfilling” crisis hypothesis. From their
analysis follows that in pursuing lower inflation, governments tolerate exchange rate
overvaluation, despite increasing probability of crisis. From here it follows that inflation
could be low and could be falling before crisis, thus having negative sign instead of positive
as we assumed at the beginning. Meanwhile, government does not have to be in the proximity
of elections in order to run low inflation and it also does not have to be a country with history
of high inflation. In case of crisis in Sweden, inflation went down before the crisis in 1992
without political elections.
Nonetheless, diverse behaviour of inflation in proximity of crisis makes it harder for
economist to make predictions, especially using the signal approach. The current nature of
such statistical tool would not be able to capture these different inflationary moments, since
we use mean and standard deviation of the whole sample.

Problems with the KLR method as estimation for inflationary rate as indicator
In fact, we would like to list the problems associated with the signal approach and then
propose possible areas to research and improve.
In terms of the signal approach itself, Berg and Patillo (1999) find the KLR method to be
having a low predicting power of crisis but it is competent at predicting “tranquil” times.
They found that 96% of crises were missed with using 50% threshold. When threshold was
lowered to 25%, the number of missed crises lowered to 74%, but the number of false signals
increased to 63%.
Mean, standard deviation – large sample
First of all, the KLR approach does not seem suitable for analysing inflation rate as indicator
of currency crisis. The methodology of the approach involves mean, standard deviation and
thresholds. In any large sample, mean and deviation are sensitive to outliers, which move
them in their direction. In our sample, countries like Argentina, Brazil, Chile, Kazakhstan and
Ukraine, experienced hyperinflation, and inflation rates would go outside 1000%. This
increased the mean, making it insensitive to small numbers, which still might experience
relatively large deviations. For example, the lowest inflation in Argentina was under 1% and
the highest 20266%. We can deduce that while KLR method might be used for variables that
do not have large variances.
Moving average
In trying to solve this problem, we took liberty using moving average method in order to try
to reduce the problem of outliers and see if predicting power increases with this method.
Similar method was used by ERW but for probit regression. We did not test all countries, just
those that experienced hyperinflation and as a sample of countries with no history of
hyperinflation, Sweden. Arbitrarily, we tried averages of different time-periods – 3 months, 6
months, 12 months, 18 months and 24 months, and 2 different coefficients – 1.5 and 2. Our
result did not show any significant improvement in signalling before crisis but there were less
“false” alarms. 12 months produced the least number of “white noise” and the coefficient of
1.5 worked well for the majority of countries in the sample – only for Argentina we used the
coefficient of 2. This is not an optimal method but it can be a benchmark for further research
in using smaller samples in order to predict currency crises, which may be applicable not only
to inflation but also to other variables.

Additionally, there exists no perfect indicator of currency crisis, so IR may not be used as a
signal by itself, but rather to be a part of Composite Index (CI). CI is sum of all indicators
that passed the threshold: CIt=Σnt=1Sit, where Sit =1 if at the time t signal was given by
indicator i. Another problem is associated with this method is that some indicators might just
pass threshold – making them “mild” indicators –while some might be initiate much stronger
signals, making them “extreme” signals. Altered CI has been proposed CIt=Σnt=1(SMit +SE it),
where SM it =1 at time t indicator gave “mild signal”, and SE it =1 if indicator gave “extreme
signal”. “Extreme” and “mild” thresholds are arbitrarily chosen by authors. In her later work,
Kaminsky (2000) finds that signals have to be weighted by their ratio of white noise-to-good
signals: CIt=Σnt=1Sit /Wt, where Wt = β/(1-α). This method may increase predicting power of
inflation but from considering our existing ratios, we can deduce that the impact on the
composite index would be minimal.

Evaluation ideas
Simple deviations from the trend in inflation have low predicting power, at least using the
KLR method, we do not exclude changes in acceleration rate could be evaluated as signals.
If IR is still to be incorporated into signal approach, mathematical methods have to be
changed to reduce the impact of outliers on the mean and standard deviation. As was
described earlier, moving averages reduce the noisiness ratio and researchers could
investigate the idea further.
While we compared ratios of two indicators, we acknowledge that exports growth rate is
considered to be one of leading indicators by literature and has more efficient ratios. To get a
better picture, comparison with other indicators, for example, bank deposits, could be done.
Also, thresholds may be adjusted to positive and negative values. Maybe, non-linear
determination of the thresholds could be tried. On the other hand, we suspect that accelerating
inflationary rate or changes in inflationary rate could be investigated as an indicator of
currency crisis since a number of countries in our sample, for example, Indonesia in 1989,
experienced accelerating inflation before the crisis. This could be a useful tool for “positive
threshold”. To comment on “negative threshold”, we believe that it could be treated the same
way as “positive threshold”, however, there is no unique mathematical method that could use
decreasing inflation as a tool for predictions.
We believe that simple numbers may be misleading and for this reason, origins of inflation
should be traced. As was illustrated before inflation could be going down because of a special
programme introduced by authorities but this programme could in fact destabilize economy
and lead to speculative attack.

Out of sample testing

We decided to run an out-sample test to see if inflationary rate will predict any crisis. Our
data set is small, thus we are limited in our analysis.
Country Crisis Good Signals False Signals EMP
India 2008m09 0 4 7.5643
Russia 2009m01 0 0 31.09421

Singapore 2009m01 0 0 8.266631

Malaysia - - 4 -
Sri Lanka - - 10 -

Limitations of the research

First of all, our analysis is based an in-sample analysis and it would be useful to run an out-
sample analysis to see how inflationary rate behaves using the signal approach.
Sample issues – conclusion issues
Our sample can be called truncated as the criteria for a country to be included in the sample
was having at least one crisis episode in the past. This restricted our analysis in terms of how
many “false” signals inflationary rate created. Furthermore, unintentionally there are more
Latin American countries, making our sample biased. Latin American countries tend to have
high inflation in general, thus some changes in inflationary rates may not be as significant as
in countries that have lower rates of inflation.
Also, our conclusion about the role of inflation does not involve many examples of crisis
episodes, since not all countries are examined. We tried to investigate at least one country in
from different regions, however, increased country coverage will definitely benefit the
analysis.
Definition of crisis
Different definition of crisis could be used. For example, Frankel and Rose (1996) define
crisis as at least 25% in the currency or at least 10% increase in the rate of depreciation.
EMP
Generally, the derivation of the crisis episodes for the analysis of inflationary rate has shown
to be problematic. We compared out crisis episodes to the episodes in the work of Edison
(2000) and while some episodes were the same, some episodes were different. Even after
using the same formula and trying the same coefficient of 2.5 for the threshold, we still did
not get the same crisis episodes as in his work. We identify a list of reasons why this could be
the case. Furthermore, the current definition of EMP itself is model-independent, meaning
none of the components of the index are derived from structural economic model. While
independence is an attractive feature, Weymark (1998) talks about how equalizing volatility
weights undermine the interpretive power and usefulness of the index as the weights in
formula that we are using, as she believes that volatility
We identify how our data set may be different mainly from Edison (2000) but our analysis is
applicable to other papers in similar field. To start with, reporting interval for indicators
varies from work to work as some authors use quarterly data, while others use monthly. Then,
variables that were chosen to represent foreign reserves and exchange rate could have been
different. On the International Financial Statistics website, we saw 13 options of official rates
of exchange and 6 options of how to represent reserves. Thus, depending on statistic chosen,
results may be manipulated. Moreover, in ERW (1996) the data is obtained from CD-ROM
version and that data was checked for transcription and other errors. We did not do this for
our analysis because of restricted abilities.
Different percentage
Furthermore, despite stating that the author was following KLR approach, in the appendix it
is written that 12-month percentage change measure was used for variables. Meanwhile, in
the original paper, reserves and exchange rate are reported as a percentage change from
previous month. We follow the original paper in this work and use percentage change from
previous month, apart from inflationary rate as it had to be done to eliminate seasonality
problem.
REER vs NEER
Then, the author does not specify if he uses nominal exchange rate or real exchange rate but
we assume the latter options, as in the appendix we can see it listed as one of the indicators.
We chose nominal exchange rate as, at first, it was used in the original paper
and secondly, we were restricted in finding data for relative consumer prices – for many
countries there were missing variables. The discrepancy between the two exchange rates can
significantly affect the results as the mathematical values will not be the same.
Hyperinflation
Also, our definition of hyperinflationary episodes and the treatment of such episodes may
differ and the author does not clarify how he dealt with the problem in his paper. However,
the problem of hyperinflation can be automatically solved if real exchange rate is used, since
it is adjusted to inflation.
Time-horizon
Finally, time-intervals vary among authors. Our sample has longer time-horizon: 1970 –
2007, while Edison (2000) from 1970 to 1995. Since the calculation of the index is
mathematical, larger samples give more information; derive different means, standard
deviation, thresholds.
Formula – interest rate
Additionally, the actual formula for the index differs between different scholars. In terms of
content of the formula, for example, in ERW (1996) they add interest rate differential as the
third determinant of the index. We did not use their formula for the same reason as stated in
the KLR paper – some countries did not have interest rate data available, so this part of the
equation had to be dropped. In terms of weights, authors dispute how to calculate them.
While we used the most popular method in the literature, in Su, Chang, Zhu and Qiao (2010)
they use inverse of standard deviation and arrive at separate list of crisis episodes.
Robustness
All of the above shows the measure for EMP is not robust to different time horizons.
Arguably, EMP may be used for severe crisis episodes while it will be insensitive to
vulnerability episodes. However, in our work is the concern that in our sample crisis episodes
could be identified wrong and thus the analysis of inflation as indicator of currency crises
could be erroneous. Furthermore, if we compare crisis episodes of Edison (2000) and Su et al.
(2010), we will see that some of the dates differ too. For example, Su et al (2010) do not list
1983 crisis in Indonesia.

Conclusion
Inflationary rate is a confusing indicator to be used in predicting crisis. Although it is a
market fundamental and fits all currency models thus logically should be included in analysis.
However, the latter could be exactly the reason why inflation is a confusing indicator. It acts
in a number of ways in approximation of speculative attack and depends on what model is
considered. If we follow first generation model, inflation accelerates as
government/authorities print out money and increase their seigniorage in order to finance
their debts, like what happened in Mexico in 1982. Or if we follow second generation model,
inflationary rate may be going down because authorities are using contractionary monetary
policy, as was demonstrated by Sweden. In third generation model, the rate can act both ways
since banking and currency crises can come in any order. In 1997 in Thailand inflationary
rate slowed down before the crisis, while in Mexico in 1987 it accelerated. Both these crises
are considered as twin crises by literature. Surprisingly, Kaminsky et al (1997) states in the
footnotes of the paper that there was one statistically significant result that showed inflation
reducing probability of an attack.
We should address the issue of why some authors conclude that domestic inflation is a
significant indicator and has an impact on probability of a crisis. In Su et al (2010) they find
IR to be significant at 5%. However, the coefficient is 0.001, meaning that if inflation
increases by 1%, probability of crisis increases by 0.001, which can be considered to have a
very low predicting power. However, in ERW work, they note the impact of inflation on the
probability is not dramatic but higher inflation increases in the odds of an attack. However,
because of non-robust results for partial correlations, authors restricted themselves from over-
interpreting their results.

Hence, we conclude that inflationary rate is not a good indicator of currency on its own and
we believe should not necessarily be included when trying to predict crisis. There are better
indicators such as exports growth and other that are proposed by the majority of literature.

Additionally, Borensztein and Gregorio (1999) investigated 41 devaluation episodes from

1970 to and they found that in their sample 30% of devaluations are followed by higher
inflation in three months. They further state that inflation rises to up to 60% after two year
and is assisted by real depreciation for longer periods. Assuming their research is right,
devaluation could be an indicator of increased inflation, rather than the other way round.

Bibliography

1. Andreou, Irène, Gilles Dufrénot, Alain Sand-Zantman, Aleksandra Zdzienicka-

Durand: “A forewarning indicator system for financial crises: the case of six Central
and Eastern European countries” 2007, Groupe d'Analyse et de Théorie Economique
(GATE), Centre national de la recherche scientifique (CNRS), Université Lyon 2,
Ecole Normale Supérieure
2. Berg, Andrew and Catherine Pattillo "Predicting currency crises:: The indicators
approach and an alternative," (1999), Journal of International Money and Finance,
Elsevier, vol. 18(4), pages 561-586, August
3. Bornsztein, Eduardo and Jose De Gregorio “Devaluation and inflation after currency
crises”1999
4. Burnside, Craig, Martin Eichenbaum and Sergio Rebelo “Currency crisis models”
2007, The New Palgrave: A Dictionary of Economics, 2nd Edition
5. Bordo, Michael D. “Currency crises (and banking crises) in historical perspective”
(1998), NBER, Research report No 10
6. Celasun, Oya “The 1994 Currency Crisis in Turkey” (1998) Macroeconomics and
Growth Group Development Research Department, World Bank Policy Research
Working Paper No. 1913
7. Cerra, Valerie and Sweta Saxena “What Caused the 1991 Currency Crisis in India?”
(2002) IMF Staff papers, Vol. 49, No. 3
8. Choi, Gongpil “An early warning system for contagious currency crises” 2002, Korea
Institute of Finance
9. Edison, Hali J. “Do indicators of financial crises work? An evaluation of an Early
Warning System” (2000)
10. Eichengreen, Barry, Andrew K. Rose, Charles Wyplosz “Contagious currency crises”
1999, CEPR Discussion paper N 1453
11. Feridun, Mete “Russian financial crisis of 1998:an econometric investigation” 2004,
International Journal of Applied Econometrics and Quantitative Studies Vol.1-4
12. Flood R. P., Marion N. P. “Perspectives on the Recent Currency Crisis Literature”
(1998) Research Department. Working Paper WP/98/130,
http://www.imforg/extemal/pubs/cat/wpl_sp.cfm?s_year=1998
13. Frankel, Jeffrey A & Rose, Andrew K, 1996. "Currency Crashes in Emerging
Markets: Empirical Indicators," (1996) CEPR Discussion Papers 1349, C.E.P.R.
Discussion PapersGoldfajn, Ilan, Rodrigo O. Valdés “Are currency crises
predictable?” European Economic Review, Volume 42, Issues 3-5, 31 May 1998,
Pages 873-885
14. Goldstein, Itay “Strategic complementarities and the Twin Crises” (2005)
15. Kalter E., Ribas A. P. “The 1994 Mexican Economic Crisis - The Role of
Government Expenditure and Relative Prices” (1999) Western Hemisphere
Department. Working Paper WP/99/160
16. Kaminsky, Graciela, “Currency and Banking Crises: The Early Warnings of Distress”
(2000)
17. Dahel, Riad “On the Predictability of Currency Crises: The Use of Indicators in the
Case of Arab Countries” 2000, Arab Planning Institute - Kuwait, Information Center
18. Kaminsky, Graciela, Saul Lizondo, Carmen M. Reinhart “Leading indicators of
currency crises” (1997)
19. Kaminsky, Graciela, Carmen M. Reinhart “The Twin Crises: The Causes of Banking
and Balance-of-Payments Problems” (1999)
20. Kindleberger, Charles P. and Robert Z. Aliber “Manias, panics and crashes” (2005)
Palgrave MacMillan
21. Krugman, Paul “The model of balance of payments crises” (1979)
22. Moore, Tomoe, Ping Wang “Can the persistence of a currency crisis be explained by
fundamentals? Markov switching models for exchange market pressure” 2009,
http://www.brunel.ac.uk/about/acad/sss/depts/economics
23. Moreno, R. (1995), “Macroeconomic Behavior During Periods of Speculative
Pressure or Realignment: Evidence from Pacific Basin Economies,” Federal Reserve
Bank of San Fransisco Working Paper, PB95-05.
24. Obstfeld, Maurice “The logic of currency crises” (1994)
25. Obstfeld, Maurice “The models of currency crises” (1995)
26. Ozatay, F.(1996), “The Lessons From the 1994 Crisis in Turkey: Public Debt
(Mis)Management and Confidence Crisis”, Yapi Kredi Economic Review, June 1996,
Vol:7, pp21-38.
27. Plata, Patricia A., Mechthild Schrooten “Mis-leading indicators? The Argentinian
Currency crisis”” (2003)
28. Sachs, Jeffrey, Aaron Tornell and Andres Velasco "The Mexican peso crisis: Sudden
death or death foretold?," (1996) Journal of International Economics, Elsevier, vol.
41(3-4), pages 265-283, November
 29. Jeffrey D. Sachs & Aaron Tornell & Andrés Velasco "Financial Crises in Emerging
Markets: The Lessons from 1995," (1996) Brookings Papers on Economic Activity,
Economic Studies Program, The Brookings Institution, vol. 27(1996-1), pages 147-
216
30. Saxena, Sweta C. “The changing nature of currency crises” 2004, Journal of
Economic Surveys Vol. 18, No. 3
31. Stein, Ernesto and Jorge Streb “Political stabilization cycles and high inflation
economies” 1997, CEMA Instituto Universitario
32. Stoker, James “Intermediation and the business cycle under a specie standard: the role
of the Gold Standard in English financial crises, 1790–1850” 1995, Mimeo,
University of Chicago
33. Su, Chi-Wei, Hsu-Ling Chang, Meng-Nan Zhu, and Zhang Qiao “An evaluation of
leading indicators of currency crises”(2010), African Journal of Business
Management Vol. 4(15), pp. 3321-3331, 4 November
34. Yeldan, Erinç “Behind the 2000/2001 Turkish crisis: Stability, Credibility, and
Governance, for Whom?” (year not specified)Bilkent University,
http://www.bilkent.edu.tr/~yeldane/Chennai_Yeldan2002.pdf
35. Velasco, A. “Financial and balance of payments crises” 1987, Journal of
Development Economics 27: 263–93.
36. http://www.econ.puc-rio.br/pdf/td420.pdf
37. http://www.assessor.ru/forum/index.php?t=789

Appendix

1. Initial sample also included Japan, Kazakhstan, Moldova, Tajikistan, Ukraine and
Kuwait but during calculations either had to be excluded from the analysis because of
missing data over a period of time either in any EMP variables or inflation rate
variables.
2. Exchange Market Pressure graphs
Argentina Bolivia
 0
5

-5
10
15

-15
-10
-120
-100
-80
-60
-40
-20
0
100

20
40
60
80
-300
-200
-100
100
200
300
400
500

0
1970m01 1970m01 1970m02
1971m08 1971m07 1971m08
1973m03 1973m01 1973m02
1974m10 1974m07 1974m08

1976m05 1976m01 1976m02

1977m12 1977m07 1977m08

1979m07 1979m01 1979m02

1980m07 1980m08
1981m02
1982m01 1982m02
1982m09
1983m07 1983m08
1984m04
1985m01 1985m02
1985m11 1986m08
1986m07
1987m06 1988m02
1988m01
1989m01

Brazil
1989m08
1989m07

Ecuador
1990m08 1991m02

Colombia
1991m01
1992m03 1992m08
1992m07
1993m10 1994m02
1994m01
1995m05 1995m08
1995m07
1997m02
1996m12 1997m01
1998m08
1998m07 1998m07
2000m02
2000m02 2000m01
2001m08
2001m09 2001m07
2003m02
2003m04 2003m01
2004m08
2004m11 2004m07 2006m02
2006m06 2006m01 2007m08
2007m07

0
5

-5
0

10
15

-20
-15
-10
200
400
600
800

0
50
-60 0
-40 0
-20 0

-50
100
150
200
250
1000
1200
1400
1600

-250
-200
-150
-100
1970m01 1970m01 1970m01
1971m08 1971m05
1971m08
1973m03 1972m09
1973m03 1974m01
1974m10 1974m10 1975m05
1976m05 1976m05 1976m09
1977m12 1977m12 1978m01
1979m07 1979m07 1979m05
1981m02 1980m09
1981m02
1982m09
1982m01
1982m09 1983m05
1984m04 1984m04 1984m09
1985m11 1985m11 1986m01
1987m06 1987m06 1987m05
Chile

1989m01 1989m01 1988m09

Finland
Denmark
1990m08 1990m01
1990m08
1991m05
1992m03 1992m03 1992m09
1993m10 1993m10 1994m01
1995m05 1995m05 1995m05
1996m12 1996m12 1996m09
1998m07 1998m07 1998m01
2000m02 1999m05
2000m02
2000m09
2001m09 2001m09
2002m01
2003m04 2003m04 2003m05
2004m11 2004m11 2004m09
2006m06 2006m06 2006m01
2007m05
 0
5

-5
10
15
20

-15
-10
0
20
40
60
80

-80
-60
-40
-20
100
120
140
0
50

-50
100

-200
-150
-100
1970m01 1971m01 1970m01
1971m08 1972m08 1971m05
1973m03 1974m03 1972m09
1974m10 1975m10
1974m01
1976m05 1977m05
1975m05
1977m12 1978m12
1976m09
1979m07 1980m07
1978m01
1981m02 1982m02
1982m09 1979m05
1983m09
1984m04 1980m09
1985m04
1985m11 1982m01
1986m11
1987m06 1983m05
1988m06
1989m01 1984m09

Jordan
1990m01
1990m08 1986m01

Indonesia
1991m08
1992m03 1987m05
1993m03
1993m10 1988m09
1994m10

Korea (Republic of)

1995m05 1990m01
1996m05
1996m12 1991m05
1997m12
1998m07 1992m09
1999m07
2000m02
2001m02 1994m01
2001m09
2002m09 1995m05
2003m04
2004m04 1996m09
2004m11
2005m11 1998m01
2006m06
2007m06 1999m05

19
93
-20
-15
-10
-5
0
5
10
15
20

-50
-40
-30
-20
-10
0
10
20

19 m1

-60
-40
-20
0
20
40
60
80
94 1 1970m01 1970m01
19 m0 1971m08
95 7 1971m04
19 m 0 1973m03 1972m07
95 3
19 m1 1974m10 1973m10
96 1
19 m0 1976m05 1975m01
97 7
1977m12
19 m 0 1976m04
97 3
1979m07
19 m1 1977m07
98 1
1981m02
19 m0 1978m10
99 7 1982m09
19 m 0 1980m01
99 3 1984m04
20 m1 1981m04
00 1 1985m11
20 m0 1982m07
01 7 1987m06
1983m10
Japan

20 m 0
01 3 1989m01

Moldova
20 m1 1990m08 1985m01

Kazakhstan
02 1
20 m0 1992m03 1986m04
03 7
20 m0 1993m10 1987m07
03 3
20 m 1 1995m05 1988m10
04 1
20 m0 1996m12 1990m01
05 7
1998m07 1991m04
20 m0
05 3 2000m02 1992m07
20 m 1
06 1 2001m09 1993m10
20 m0
07 7 2003m04 1995m01
20 m0
07 3 2004m11 1996m04
m
11 2006m06
1997m07
1998m10
 -80
-60
-40
-20
0
20
40
60
80

-30
-20
-10
0
10
20
30
40
50
-30
-20
-10
0
10
20
30
40
50
60
1970m01
1970m01 1970m01
1971m08 1971m09
1971m09
1973m03 1973m05 1973m05
1974m10 1975m01 1975m01
1976m05 1976m09 1976m09
1977m12 1978m05 1978m05
1979m07 1980m01 1980m01
1981m02 1981m09 1981m09
1982m09 1983m05 1983m05
1984m04
1985m01 1985m01
1985m11
1986m09 1986m09
1987m06
1988m05 1988m05
1989m01
1990m01 1990m01

Paraguay
1990m08

Singapore
1991m09

Philippines
1991m09
1992m03
1993m05 1993m05
1993m10
1995m01 1995m01
1995m05
1996m09 1996m09
1996m12
1998m05 1998m05
1998m07
2000m01
2000m02 2000m01
2001m09
2001m09 2001m09
2003m05
2003m04 2003m05
2005m01
2004m11 2005m01
2006m06 2006m09
2006m09

-300
-200
-100
100
200
300
400
500
600
700

0
20
40
60
80
100
120

-20
1994m12 1970m01
1971m09
-6000
-4000
-2000
2000
4000
6000
8000

1995m07
1992m03
1996m02 1973m05
1992m12
1996m09 1975m01
1993m09
1997m04 1976m09
1994m06
1997m11 1978m05
1995m03
1998m06 1980m01
1995m12
1999m01 1981m09
1996m09
1999m08 1983m05
1997m06
2000m03 1985m01 1998m03
1986m09 1998m12
2000m10
Peru

2001m05 1988m05 1999m09

Russia

Sweden
1990m01 2000m06
2001m12
1991m09 2001m03
2002m07
1993m05 2001m12
2003m02
1995m01 2002m09
2003m09
1996m09 2003m06
2004m04
1998m05 2004m03
2004m11
2000m01 2004m12
2005m06
2005m09
2001m09
2006m01
2006m06
2003m05
2006m08 2007m03
2005m01
2007m03 2007m12
2006m09
2007m10
 0
5

-5
10
15

-50
0
50
100
150
200
250
300
-2 0
-1 5
-1 0

-600
-400
-200
0
200
400
600
800
1000
1200
1992m12 1970m01
1996m04
1993m08 1971m06
1996m10
1994m04 1972m11
1997m04
1994m12
1974m04
1997m10
1995m08
1975m09
1998m04
1996m04
1977m02
1998m10
1996m12
1978m07
1999m04 1979m12
1997m08
1999m10 1981m05
1998m04
2000m04 1982m10
1998m12
2000m10 1984m03
1999m08
2001m04 1985m08
2000m04
2001m10 1987m01

Ukraine

Uruguay
2000m12
2002m04

Tajikistan
1988m06
2001m08
2002m10 1989m11
2002m04 2003m04 1991m04
2002m12 2003m10 1992m09
2003m08 2004m04 1994m02
2004m04 2004m10 1995m07
2004m12 2005m04 1996m12
2005m08 2005m10 1998m05
2006m04 2006m04 1999m10
2006m12 2006m10 2001m03
2007m08 2007m04 2002m08
2007m10 2004m01
-20
-10
10
20
30
40

2005m06

-30
-20
-10
0
10
20
30

1970m01

-40
-30
-20
-10
0
10
20
1970m01 2006m11
1970m01 1971m09
1971m08
1971m08 1973m05
1973m03
1973m03
1974m10 1975m01
1974m10
1976m05 1976m09
1976m05
1977m12 1978m05
1977m12
1979m07 1979m07 1980m01
1981m02 1981m02 1981m09
1982m09 1982m09 1983m05
1984m04 1984m04
1985m01
1985m11 1985m11
1986m09
1987m06 1987m06

UK
1988m05
1989m01 1989m01

Kuwait
1990m01
Thailand

1990m08 1990m08
1992m03
1991m09
1992m03
1993m10 1993m05
1993m10
1995m05 1995m01
1995m05
1996m12 1996m09
1996m12
1998m07
1998m07 1998m05
2000m02
2000m02 2000m01
2001m09
2001m09 2001m09
2003m04
2003m04 2003m05
2004m11
2006m06 2004m11 2005m01
2006m06 2006m09
 0
5

-5
10
15
20
25

-10

-20
0
20
40
60
80
-60
-40
-20
0
20
40
60
80
100
120

100
120
1970m01
1970m01 1970m01
1971m09
1971m08 1971m08
1973m03
1973m05 1973m03
1974m10 1975m01 1974m10
1976m05 1976m09 1976m05
1977m12 1978m05 1977m12
1979m07 1980m01 1979m07
1981m02 1981m09 1981m02
1982m09 1983m05 1982m09
1984m04 1985m01 1984m04
1985m11 1985m11
1986m09
1987m06 1987m06
1988m05

India
1989m01 1989m01

Turkey
1990m01
1990m08 1990m08

Venezuela
1991m09
1992m03 1992m03
1993m05
1993m10 1993m10
1995m01
1995m05 1995m05
1996m12 1996m09 1996m12
1998m07 1998m05 1998m07
2000m02 2000m01 2000m02
2001m09 2001m09 2001m09
2003m04 2003m05 2003m04
2004m11 2005m01 2004m11
2006m06 2006m09 2006m06

-25
-20
-15
-10
10
15
20
25

0
5

-5
0
20
40
60
80

-60
-40
-20
100
0
2
4
6
8

-8
-6
-4
-2

-14
-12
-10

1970m01 1970m01 1970m01

1971m09 1971m09 1971m06
1973m05 1973m05 1972m11
1975m01 1975m01 1974m04
1976m09 1976m09 1975m09
1978m05 1978m05 1977m02
1980m01 1980m01 1978m07
1981m09 1981m09 1979m12
1983m05 1983m05 1981m05
1985m01 1985m01 1982m10
1986m09 1986m09 1984m03
1988m05 1988m05 1985m08
1990m01 1990m01
1987m01
Malaysia
Sri Lanka

1991m09
1988m06
1991m09
1989m11
1993m05 1993m05
1991m04
1995m01 1995m01
1992m09
1996m09 1996m09
1994m02
1998m05 1998m05
1995m07
2000m01 2000m01
1996m12
2001m09 2001m09
1998m05
2003m05 2003m05
1999m10
2005m01 2005m01 2001m03
2006m09 2006m09 2002m08
2004m01
2005m06
2006m11
-40
-20
0
20
40
60
80
100
120
140
160
180
1970m01
1971m08
1973m03
1974m10
1976m05
1977m12
1979m07
1981m02
1982m09
1984m04
1985m11
1987m06
1989m01
1990m08
1992m03
1993m10
1995m05
1996m12
1998m07
2000m02
2001m09
2003m04
2004m11
2006m06