2. Literature Review
2.2 Variables for Predicting Financial Crisis and Economic Recession
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
9
Fourth, Connors has proved that VIX is able to tell when market top or bottom is.
2.2 Variables for Predicting Financial Crisis and Economic Recession
Next, this paper reviews some literatures about the issue of crisis prediction and generalizes the economic and financial indicators to identify macroeconomic condition, predict occurrence of crisis, or construct the earning warning systems.
Berg and Pattillo (1999) use two kinds of models and a number of macro variables as predictors to examine whether the model with macroeconomic variables would have helped predict the 1997 currency crises. They have shown that the bilateral real exchange rate, reserve growth, export growth, growth of M2/ reserves, current account deficit and ratio of M2 to reserves are important risk factors for the prediction of currency crises.
Hatzius et al. (2010) build a new financial condition indexes (FCIs) which includes not only interest rates and asset prices with longer data periods, but a broad range of quantitative and survey-based indicators. Take some of total 45 indicators for example, 10-Year treasury note yield at constant maturity, 10yr T-note/3month T-bill spread, Baa/10yr T-note spread, monthly average VIX, financial market capitalization, real broad trade-weighted exchange value of the USD, banks CDS spread, and price of oil are the components. They find the overall index performed better than any of its major components in recent years.
Duca and Peltonen (2011) use quarterly data of 28 emerging market and advanced economies since 1990 to develop a framework to assess systematic risks and to predict periods of extreme financial instability. They test the ability of a wide range of composite indicators such as asset price (equity and property prices), credit (credit and monetary aggregates) and macro developments (GDP, inflation,
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
10
government deficit, current account deficit). They also construct a financial stress index which is a country-specific composite index, and the five components are:
(1) the spread of the 3-month interbank rate over the 3-month Government bill rate;
(2) negative quarterly equity returns;
(3) the realized volatility of the main equity index;
(4) the realized volatility of the nominal effective exchange rate;
(5) the realized volatility of the yield on the 3-month Government bill.
They find that considering jointly various indicators in a multivariate framework and taking into account jointly domestic and global macro-financial vulnerabilities greatly improves the performance of the models in forecasting systemic events.
Bailey et al. (2012) look for the factors of one-minute changes in VIX to understand whether the risk neutral volatility and risk premium are affected by more forces. They use the explanatory variables based on trade and quote information including SPY trading volume, the price-setting or aggressive buy-sell imbalance of SPY, the bid-ask spread of SPY, trading volume of GLD, buy-sell imbalance of GLD, and changes in bid-ask spreads for the CDX NAIG index which reflects both corporate default risk and bond market liquidity. They find that a significant portion of VIX variability relates to trader behavior and macroeconomic fundamentals, so macroeconomic influences are significant. Temporary price effects are also observed around macroeconomic news releases.
Soylemez (2012) tests for the relevant direction of Granger causality between the VIX index and S&P500 returns, and the results show that S&P500 influences VIX, but the opposite relationship does not exist.
Jan Babecký et al. (2012) construct a dataset covering 36 developed countries from 1970 to 2010 at quarterly frequency to examine which potential leading indicators preceding economic crises are most useful in explaining the developed
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
11
economy’s real costs resulting from crises. They identified over 100 relevant macroeconomic and financial variables based on past studies and finally choose 30 variables as predictors. They find that about a third of the potential early warning indicators are useful for explaining the incidence of economic crises in EU and OECD countries in the past 40 years. The key early warning signal is growth in domestic credit to the private sector, while increase in government debt, the current account deficit, FDI inflow, or a fall in house prices and share prices could be considered late early warning indicators. The 30 predictors they choose are below.
Table 1. Predictors in Jan Babecký et al. (2012)
No. Variable No. Variable
1 BAA corporate bond spread 16 M3 2 Gross total fixed capital formation
(constant prices) 17 Money market interest rate 3 Commodity prices 18 Nominal effective exchange rate 4 Current account (%GDP) 19 Net national savings (%GNI) 5 Domestic credit to private sector
(%GDP) 20 Stock market index
6 FDI net inflows (%GDP) 21 Total tax burden (%GDP) 7 Government consumption (constant
prices) 22 Terms of trade
8 Government debt (%GDP) 23 Trade (%GDP) 9 Private final consumption expenditure
(constant prices) 24 Trade balance
10 Gross liabilities of personal sector 25 Global domestic credit to private sector (%GDP)
11 House price index 26 Global FDI inflow (%GDP) 12 Industrial production index 27 Global inflation
13 Industry share (%GDP) 28 Global GDP
14 Consumer price index 29 Global trade (constant prices)
15 M1 30 Long term bond yield – money market
interest rate
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
12
This paper also reviews some studies which apply the Markov regime switching model to identify the state switching in macroeconomic time series data. The origin and evolution of the model will be covered in the next part, but now this paper introduces some recent literatures about the application of Markov switching model first, especially focuses on the economic or financial variables using in these studies, and generalizes their influences to the models.
Arias and Erlandsson (2005) propose an innovative early warning system of currency crises modeled by a time-varying Markov regime switching model applied to six emerging South-East Asian countries. The sample covers the period from 1989 to 2002 on a monthly basis, and the explanatory variables belonging to various economic categories such as public deficit, central bank's credit to the public sector, production growth, inflation, stock prices, banking fragility, trade balance deficit, the real exchange rate, capital account, and 3-months LIBOR interest rate are chosen to enter the transition probability equation. The forecasting performance is satisfactory, so they believe the time-varying transition probability Markov regime switching models are more adequate to build early warning systems of currency crises.
Chen (2009) investigates whether macro variables were useful in predicting bear markets, where bear market is defined as the regime with low return and high volatility of the S&P 500 price index. He focuses on the U.S. stock market and investigates the S&P 500 price index return from February, 1957 to December, 2007 using the monthly returns. Various macro variables include yield spreads (the difference between the 3-Month Treasury Bill Rate and the 10-Year Treasury Constant Maturity Rate, and the difference between the 3-Month Treasury Bill Rate and the 5-Year Treasury Constant Maturity Rate), inflation rates (consumer prices), money stocks (M1 and M2), aggregate output (industrial production), unemployment rates, federal funds rates, nominal effective exchange rates, and federal government
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
13
debts. He uses a modified version of the two-state Markov-switching model to examine empirically the cyclical variations in stock returns. Empirical evidence identifies there are two states in the S&P 500 stock index, the high-return stable and low-return volatile states in stock returns are conventionally labeled as bull markets and bear markets, respectively. He also finds term spreads and inflation rates are the most useful and significant predictors of recessions in the U.S. stock market.
Mulvey and Zhao (2010) develop a dynamic approach to capture the critical market sentiments, with expected asset returns highly dependent on the associated economic regimes. Expected equity returns are characterized by a set of eight economic factors within a regime-switching auto-regressive approach, including changes in S&P 500 price index, changes in Treasury bond yields, changes in U.S.
dollar index, changes in implied volatility, changes in aggregate dividend yield, short term interest rate, changes in treasury yield spread, and changes in credit spread. For constructing a portfolio with the Markov regime switching model, they employ exchange traded funds to test the approach. Their findings are alternative regimes lead to differing correlation and return expectations over the designated asset classes, thus the developed investment portfolio provides much higher returns with less risk than equity indices during the period of January 1999 to October 2010.
Baba and Sakurai (2011) use the Markov regime switching approach to investigate which macroeconomic variables can predict regime shifts in the VIX index. They choose the monthly VIX index as the dependent variable, and the following seven macroeconomic variables to test their influence on the switching probabilities from one regime to another: Consumer Price Index, Producer Price Index of finished goods, manufacturing capacity utilization, industrial production, the Federal funds rate, the difference between 5-year US government yield over 3-month Treasury bill rate (TERM 5Yt) and the difference between 10-year US government
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
14
yield over 3-month Treasury bill rate (TERM 10Yt). The two term spreads, TERM5Yt and TERM 10Yt, are found to have statistically significant positive coefficients in predicting the regime shifts from tranquil to turmoil regime. The likelihood ratio test also supports the models with each term spread have a significantly higher explanatory power compared with the baseline model without a macroeconomic variable.
Engemann et al. (2011) consider whether oil price shocks significantly increase the probability that the economy enters recession in some industrialized countries.
They adopt a Markov switching model in which the transition probabilities vary with both net oil shocks and the term spread defined as the 10-year and 3-month Treasury equivalents for each country to estimate the turning points. They find that oil shocks indeed affect the probability that the economy transfers from expansion to recession for most countries.
To sum up, Mulvey and Zhao (2010) say that financial markets are usually characterized as bullish or bearish, based on the economic indicators. It is observed that the equity market returns and volatility move in opposite directions more than 70% of the time. Credit spreads and yield spreads are generally wider in a downward market than that in an upward market. Stock markets may stay in one state for some time before moving to another state at a later date. It is highly probable that market sentiment, market volatility, and the non-smooth asset return processes are state dependent. Therefore, ignoring such a possibility and information across the market regimes may cause incomplete investment strategies. A regime switching process is to characterize the sudden movements in market driven by the market information.
‧
Table 2. Summary of indicators
Authors Main idea of study Indicators using in paper Berg and Pattillo (1999) using two kinds of models and a
number of macro variables as predictors to examine whether the model with macroeconomic variables would have helped predict the 1997 currency crises.
(1) real exchange rate (2) reserve growth (3) export growth
(4) growth of M2/reserves (5) current account deficit (6) ratio of M2 to reserves Hatzius et al. (2010) building a new financial
condition indexes (FCIs) which includes longer data history, and a broad range of quantitative and survey-based indicators. They find the new index performed better in recent years than any of its major components.
(1) 10-Year treasury note yield at constant maturity
(2) 10yr T-note/3month T-bill spread
(3) Baa/10yr T-note spread (4) monthly average VIX (5) financial market capitalization
(6) real broad trade-weighted exchange value of the USD (7) banks CDS spread (8) price of oil Duca and Peltonen
(2011)
using data of 28 emerging market and advanced economies since 1990 to develop a
framework to assess systematic risks and to predict periods of extreme financial instability.
They also construct a financial stress index which is a
country-specific composite index.
(1) the spread of the 3-month interbank rate over the 3-month Government bill rate
(2) negative quarterly equity returns
(3) the realized volatility of the main equity index
(4) the realized volatility of the nominal effective exchange rate
(5) the realized volatility of the yield on the 3-month
Government bill Bailey et al. (2012) seeking the roots of one-minute
changes in VIX to understand whether the risk neutral volatility and its risk premium are affected by more forces.
(1) SPY trading volume (2) the price-setting or
aggressive buy-sell imbalance of SPY
(3) the bid-ask spread of SPY
‧
They find a significant portion of VIX variability relates to trader behavior and
macroeconomic fundamentals.
(4) trading volume of GLD (5) buy-sell imbalance of GLD (6) changes in bid-ask spreads for the CDX NAIG index Jan Babecký et al.
(2012)
constructing a dataset of 36 developed countries from 1970 to 2010 to examine which potential leading indicators preceding economic crises are most useful in explaining the developed economy’s real costs resulting from crises.
Quarterly frequency:
See Table 1.
Arias and Erlandsson (2005)
proposing an innovative early warning system of currency crises modeled by a
time-varying Markov regime switching model applied to six emerging South-East Asian countries.
Monthly basis:
(1) public deficit
(2) central bank's credit to the public sector
(3) production growth (4) inflation
(5) stock prices (6) banking fragility (7) trade balance deficit (8) the real exchange rate (9) capital account (10) 3-month LIBOR rate Chen (2009) investigating whether macro
variables were useful in predicting bear markets from February, 1957 to December, 2007. He uses the two-state Markov-switching model to examine empirically the cyclical variations, and identifies there are two states in the S&P 500 stock index.
Monthly returns:
(1) 10-year yield spreads (2) 5-year yield spreads (3) inflation rates (4) M1 and M2
(5) industrial production (6) unemployment rates (7) federal funds rates
(8) nominal effective exchange rates
(9) federal government debts Mulvey and Zhao
(2010)
developing a dynamic
systematic approach to capture the critical market sentiments,
(1) changes in S&P 500 price index
(2) changes in Treasury bond
‧
with expected asset returns highly dependent on the associated economic regimes.yields
(3) changes in U.S. dollar index (4) changes in implied
volatility
(5) changes in aggregate dividend yield
(6) short term interest rate (7) changes in treasury yield spread
(8) changes in credit spread Baba and Sakurai
(2011)
using the Markov regime switching approach to investigate which
macroeconomic variables can predict regime shifts in the VIX index. The two term spreads are found to have statistically significant positive coefficients in predicting the regime shifts
Monthly basis:
(1) Consumer Price Index (2) Producer Price Index of finished goods
(3) manufacturing capacity utilization
(4) industrial production (5) the Federal funds rate (6) the difference between 5-year US government yield over 3-month Treasury bill rate (7) the difference between 10-year US government yield over 3-month Treasury bill rate Engemann et al. (2011) considering whether oil price
shocks significantly increase the probability that the economy enters recession in some industrialized countries. They adopt a Markov switching model in which the transition probabilities vary to estimate the turning points.
(1) net oil shocks
(2) the term spread defined as the 10-year and 3-month Treasury equivalents
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
18