Research article

Simple is best: A single-CNN method for classifying remote sensing images

  • Received: 18 June 2023 Revised: 13 August 2023 Accepted: 22 August 2023 Published: 05 September 2023
  • Recently, researchers have proposed a lot of methods to boost the performance of convolutional neural networks (CNNs) for classifying remote sensing images (RSI). However, the methods' performance improvements were insignificant, while time and hardware costs increased dramatically due to re-modeling. To tackle this problem, this study sought a simple, lightweight, yet more accurate solution for RSI semantic classification (RSI-SC). At first, we proposed a set of mathematical derivations to analyze and identify the best way among different technical roadmaps. Afterward, we selected a simple route that can significantly boost a single CNN's performance while maintaining simplicity and reducing costs in time and hardware. The proposed method, called RE-EfficientNet, only consists of a lightweight EfficientNet-B3 and a concise training algorithm named RE-CNN. The novelty of RE-EfficientNet and RE-CNN includes the following: First, EfficientNet-B3 employs transfer learning from ImageNet-1K and excludes any complicated re-modeling. It can adequately utilize the easily accessible pre-trained weights for time savings and avoid the pre-training effect being weakened due to re-modeling. Second, RE-CNN includes an effective combination of data augmentation (DA) transformations and two modified training tricks (TTs). It can alleviate the data distribution shift from DA-processed training sets and make the TTs more effective through modification according to the inherent nature of RSI. Extensive experimental results on two RSI sets prove that RE-EfficientNet can surpass all 30 cutting-edge methods published before 2023. It gives a remarkable improvement of 0.50% to 0.75% in overall accuracy (OA) and a 75% or more reduction in parameters. The ablation experiment also reveals that RE-CNN can improve CNN OA by 0.55% to 1.10%. All the results indicate that RE-EfficientNet is a simple, lightweight and more accurate solution for RSI-SC. In addition, we argue that the ideas proposed in this work about how to choose an appropriate model and training algorithm can help us find more efficient approaches in the future.

    Citation: Huaxiang Song, Yong Zhou. Simple is best: A single-CNN method for classifying remote sensing images[J]. Networks and Heterogeneous Media, 2023, 18(4): 1600-1629. doi: 10.3934/nhm.2023070

    Related Papers:

    [1] Yuan-Ming Lee, Kuan-Min Wang . How do Economic Growth Asymmetry and Inflation Expectations Affect Fisher Hypothesis and Fama’s Proxy Hypothesis?. Quantitative Finance and Economics, 2017, 1(4): 428-453. doi: 10.3934/QFE.2017.4.428
    [2] Thomas C. Chiang . Economic policy uncertainty and stock returns—evidence from the Japanese market. Quantitative Finance and Economics, 2020, 4(3): 430-458. doi: 10.3934/QFE.2020020
    [3] Md Akther Uddin, Mohammad Enamul Hoque, Md Hakim Ali . International economic policy uncertainty and stock market returns of Bangladesh: evidence from linear and nonlinear model. Quantitative Finance and Economics, 2020, 4(2): 236-251. doi: 10.3934/QFE.2020011
    [4] Lucrezia Rosini, Vijay Shenai . Stock returns and calendar anomalies on the London Stock Exchange in the dynamic perspective of the Adaptive Market Hypothesis: A study of FTSE100 & FTSE250 indices over a ten year period. Quantitative Finance and Economics, 2020, 4(1): 121-147. doi: 10.3934/QFE.2020006
    [5] Haryo Kuncoro, Fafurida Fafurida, Izaan Azyan Bin Abdul Jamil . Growth volatility in the inflation-targeting regime: Evidence from Indonesia. Quantitative Finance and Economics, 2024, 8(2): 235-254. doi: 10.3934/QFE.2024009
    [6] Andrey Kudryavtsev . Absolute Stock Returns and Trading Volumes: Psychological Insights. Quantitative Finance and Economics, 2017, 1(2): 186-204. doi: 10.3934/QFE.2017.2.186
    [7] Raéf Bahrini, Assaf Filfilan . Impact of the novel coronavirus on stock market returns: evidence from GCC countries. Quantitative Finance and Economics, 2020, 4(4): 640-652. doi: 10.3934/QFE.2020029
    [8] Arifenur Güngör, Hüseyin Taştan . On macroeconomic determinants of co-movements among international stock markets: evidence from DCC-MIDAS approach. Quantitative Finance and Economics, 2021, 5(1): 19-39. doi: 10.3934/QFE.2021002
    [9] Diogo Matos, Luís Pacheco, Júlio Lobão . Availability heuristic and reversals following large stock price changes: evidence from the FTSE 100. Quantitative Finance and Economics, 2022, 6(1): 54-82. doi: 10.3934/QFE.2022003
    [10] Mohsen Bahmani-Oskooee, Sujata Saha . On the effects of policy uncertainty on stock prices: an asymmetric analysis. Quantitative Finance and Economics, 2019, 3(2): 412-424. doi: 10.3934/QFE.2019.2.412
  • Recently, researchers have proposed a lot of methods to boost the performance of convolutional neural networks (CNNs) for classifying remote sensing images (RSI). However, the methods' performance improvements were insignificant, while time and hardware costs increased dramatically due to re-modeling. To tackle this problem, this study sought a simple, lightweight, yet more accurate solution for RSI semantic classification (RSI-SC). At first, we proposed a set of mathematical derivations to analyze and identify the best way among different technical roadmaps. Afterward, we selected a simple route that can significantly boost a single CNN's performance while maintaining simplicity and reducing costs in time and hardware. The proposed method, called RE-EfficientNet, only consists of a lightweight EfficientNet-B3 and a concise training algorithm named RE-CNN. The novelty of RE-EfficientNet and RE-CNN includes the following: First, EfficientNet-B3 employs transfer learning from ImageNet-1K and excludes any complicated re-modeling. It can adequately utilize the easily accessible pre-trained weights for time savings and avoid the pre-training effect being weakened due to re-modeling. Second, RE-CNN includes an effective combination of data augmentation (DA) transformations and two modified training tricks (TTs). It can alleviate the data distribution shift from DA-processed training sets and make the TTs more effective through modification according to the inherent nature of RSI. Extensive experimental results on two RSI sets prove that RE-EfficientNet can surpass all 30 cutting-edge methods published before 2023. It gives a remarkable improvement of 0.50% to 0.75% in overall accuracy (OA) and a 75% or more reduction in parameters. The ablation experiment also reveals that RE-CNN can improve CNN OA by 0.55% to 1.10%. All the results indicate that RE-EfficientNet is a simple, lightweight and more accurate solution for RSI-SC. In addition, we argue that the ideas proposed in this work about how to choose an appropriate model and training algorithm can help us find more efficient approaches in the future.



    Purpose: We present an uncertainty hypothesis, which provides evidence for examining the relation between stock returns and (expected) inflation based on the data of 20 major global stock markets.

    Design/methodology/approach: The GED-APARCH model is used to estimate the relation between stock market returns and inflation while controlling for extreme shocks arising from the 2008–2009 global financial crisis (GFC) and 2019–2020 COVID-19 pandemic.

    Findings: Based on the testing sample period from January 1990 to June 2023 in 20 major countries, we found strong evidence to support a negative correlation between market returns and (expected) inflation in 18 out of the 20 countries, thereby confirming studies by Fama (1981), Geske and Roll (1983). Further evidence reveals that EMV is positively associated with inflation. This evidence shows that EMV produces a negative effect on U.S. as well as global stock returns supporting the uncertainty hypothesis (Chiang, 2023) and the U.S. volatility spillover hypothesis. We conclude that both domestic inflation and U.S. equity market volatility (EMV) have harmful effects on stock returns. Ignoring the significance of U.S. EMV inevitably produces a biased estimator in the test equation. The current study is robust across different specifications of inflation and stock market returns.

    Research limitations/implications: A different model such as a VAR approach can be used to examine the null hypothesis. The variables included in the VAR may include (real) stock market returns, expected inflation, expected real output and monetary growth. However, this extension is more appropriately explored in another paper.

    Practical implications: This study has contributed to the practical implications for portfolio management. Essentially, this study identifies two major risk variables, EMV due to inflation and EMV associated with interest rate changes. The former captures investors' sentiment, whereas the latter highlights the Fed's policy stance. Both variables exhibit a negative effect on stock returns.

    Originality/value: This study develops a new model to explain the negative relation between stock market returns and (expected) inflation. The link is through the U.S. EMV, which is positively associated with heightened inflation or an upward shift in the Fed's policy but negatively affects stock returns. In addition to providing an empirical estimation, this study employs an asymmetric power GARCH model, which is rich in modeling various properties of heteroskedasticity and long memory.

    For the past four decades, the Fed has been able to keep the inflation rate around its 2% target level largely because of its effective management of its monetary policy. These decades of rate stability ended as a consequence of the massive quantitative easing implemented as part of the rescue plans aimed at relieving the COVID-19 pandemic. In fact, the inflation rate surged, reaching 8.2% as reported by the U.S. Bureau of Labor Statistics on September 13, 2022. In an effort to combat the rampant inflation, the Fed started to raise interest rates in mid-March 2022 and has hiked interest rates at 11 consecutive meetings up to the FOMC meeting on July 25–26 2023, when the fed funds rate was in the range of 5.25–5.50%, the highest level in more than 22 years. The need for further rate hikes and policy uncertainty depend on the prevailing future inflation rate. In retrospect, the Fed's determination to fight inflation caused a significant selloff in stocks, a result of investors' fears and ongoing pessimistic economic outlook, which could further depress the economy as it results in a recession. Viewed from this perspective, there exists a prospect that stock market returns are negatively associated with inflation expectations, without even accounting for the effects of a rise in geopolitical risks (Chiang, 2021; Caldara and Iacoviello, 2022) resulting from the Russian-Ukraine conflicts and its impact on inflation in 2022–2023.

    Evidence of the negative relation between stock market returns and inflation expectation was documented by Fama (1981). In his penetrating 1981 article, Fama proposes a "proxy hypothesis" and maintains that a rise in stock returns signals an increase in output; however, output increase is negatively associated with a rise in inflation. Gallagher and Taylor (2002) test U.S. data and find results to support the proxy hypothesis. Yet, Kaul (1987) finds evidence that the negative relation between stock market returns and inflation expectation in the U.S. data can be explained only in times of a counter-cyclical monetary policy. In addition, Liu et al. (1993) report that there is no significant connection between real stock market returns and expected real economic activity. Yet, the finding of a negative correlation between real stock market returns and real economic performance by Majid (2010) further casts some doubt on Fama's underlying connections.

    Geske and Roll (1983) propose a reverse causation relation, arguing that a drop in stock market prices reveals a decline in economic performance and potential tax revenue; this decline in tax revenue will cause larger government budget deficits financed by monetization that result in higher inflation expectations. James et al. (1985) examine the relation among different macroeconomic variables, including stock market returns, real economic activity, inflation and monetary growth, using a multivariate time series model and find sound evidence to back up Geske and Roll's (1983) hypothesis.

    Neither Fama's proxy hypothesis that presumes a negative relation between stock market returns and inflation expectations nor Geske and Roll's (1983) causal hypothesis, which starts with changes in real activity that induces monetary/fiscal policy response and in turn a change in inflation, can be used to fully illustrate the negative stock market and inflation relation. To stomp out the recent surge in inflation, the Fed committee has actively raised interest rates. The uncertainty of future rate hikes, however, has created fears that have increased EMV and, in turn, jeopardized stock market performance. We propose an uncertainty hypothesis that links the relation of inflation-equity market volatility and stock returns. The evidence thus provides insights into the uncertainty approach to validate investors' sentiment as well as the spillover effect from the Fed's actions to U.S. and subsequently international stock market performances.

    In conducting empirical estimations, we use a sample of 20 major countries to examine the relation between stock returns and inflation. This broad sample of countries helps not only to reduce the biasedness from using a small set of countries but also provides better information with respect to different countries' behavior.

    This paper makes several contributions. First, this study employs an asymmetrical power GARCCH model (APARCH), which covers the various features of heteroskedasticity, including asymmetry, variance clustering and long-memory process. In addition, the model includes dummy variables to control for capturing impacts arising from 2008–2009 global financial crisis and 2020–2021 COVID-19 pandemic. Second, this study shows that stock market returns move positively with conditional variance and confirms the risk-return trade-off hypothesis (Bali and Engle, 2010). Third, there are strong negative correlations between stock returns and (expected) inflation for 18 out of 20 major markets; the exceptions are Brazil and Russia, which are positively correlated due to higher prevailing inflation. Fourth, evidence shows that inflation in U.S. and EMV (both inflation and monetary policy induced) are significantly and positively correlated, and EMV has a significant negative effect on U.S. and global stock returns. This volatility-stock return effect adds an additional channel that explains stock returns. Fifth, besides domestic inflation, the news of changes in EMV has an incremental power to explain stock return equations. Thus, the U.S. stock market risk should be explicitly priced into stocks for each country's investments.

    The remainder of this paper is organized as follows. Section 2 contains a literature review. Section 3 outlines an econometric model, which features a GED-APGARCH model. Section 4 provides a description of the data for empirical estimations. Section 5 presents empirical results for the stock market return equation. Section 6 reports robustness tests for the real stock return equation. Section 7 discusses conclusions for the empirical findings.

    The Fisher hypothesis (1930) posits that the stock market returns are positively related to the expected inflation. The rationale for this behavior stems from the notion that compensation is required to offset a loss from inflation. To test the Fisher hypothesis, Boudoukh and Richardson (1993) use long historical data to investigate markets in the U.S. and U.K. and find sound evidence to support the Fisher hypothesis. In a subsequent study using multi-country panel data, Solnik and Solnik (1997) reach comparable results that are in line with the early results obtained by Boudoukh and Richardson (1993) and conclude that the Fisher hypothesis is valid in the long run. Wong and Wu (2003) examine data for G7 and eight Asian countries and find a positive relationship for the Fisher hypothesis in a long horizon. Hasan (2008) examines the Fisher hypothesis using U.K. data. Toyoshima and Hamori (2011) investigate this issue by adding data for the U.S. and Japan, and their testing results agree with the Fisher hypothesis and the existence of a positive relation between stock market returns and inflation. This evidence implies that stocks can be viewed as a good instrument for hedging against inflation.

    However, a number of studies reject the Fisher hypothesis. For instance, Nelson (1976) investigates U.S. data and finds evidence that fails to support the Fisher hypothesis. His results reveal a negative relation between stock returns and an expected change in the inflation rate. By the same token, Gultekin (1983) estimates the data for 26 countries and fails to find consistent support for the Fisher hypothesis in most countries. Davis and Kutan (2003) and Spyrou (2004) examine the Fisher effect on international markets; however, they find very little evidence to support the Fisher hypothesis.

    Several studies have shown mixed outcomes depending on the sources of shocks, sample periods and monetary regimes used in their analyses. For instance, Marshall (1992) finds that stock market returns are negatively correlated with expected inflation when inflation is generated by fluctuations in real economic activity. Cifter (2015) examines data for Mexico and South Africa and reports that the coefficient of real stock return-inflation relation is negative during periods of recession. Narayan and Zheng (2010), who investigate the U.K. stock market, report that stocks lack the ability to hedge against inflation in the short horizon. Tiwari et al. (2022) use long historical, annual U.S. data to investigate the relation between stock market returns and inflation. The research using wavelet techniques shows evidence of a weak co-movement between stock returns and inflation in the short run and a stronger co-movement toward the long run.

    Al-khazali (2003) investigates data for a group of emerging markets and finds a negative relation between stock market prices and inflation in all countries, except Malaysia. Graham (1996) tests the stock return in relation to inflation under different monetary policy regimes. His results indicate the presence of a negative relation in times when monetary policy is implemented during a counter-cyclical or neutral period; a positive relation occurs when the policy is conducted in a procyclical phase.1

    1The intention of this section is to provide a review of major studies rather than an exhaustive review of literature. Madadpour and Asgari (2019) provide a more detailed literature review of early studies.

    In their earlier study, Modigliani and Cohn (1979) suggest a money illusion hypothesis and contend a rational investor should anticipate that cash flows would move with the inflation rate, rising in proportion to the discount rate. However, if investors fail to adjust equity cash flows upward in the same proportion as the upward shift in the discount rate during higher expected inflation, the undervaluation in stocks will lead to a drop in the stock price. Testing of NYSE, AMEX and NASDAQ stocks, evidence from Cohen et al. (2005) supports the model with money illusion.

    Fama's (1981) proxy hypothesis is pertinent to explain the negative relation between real stock returns and inflation. Fama's idea is consistent with the studies by Nelson (1976), Fama and Schwert (1977). Fama (1981) contends that there is a positive relation between real stock returns and real economic activity and a negative relation between real economic activity and expected inflation, which results in a negative relation between real stock returns and expected inflation. Gallagher and Taylor (2002) conduct an empirical analysis that confirms the proxy hypothesis. In testing the data for the U.S., U.K., Germany and Canada, Liu et al. (1993) demonstrate that expected inflation is negatively correlated with expected real production; however, they found no significant evidence to support the relation between real stock returns and expected real production, which leads to a rejection of the proxy hypothesis. In his study of Indonesian data, Majid (2010) fails to find evidence to support the inverse relation between real stock market returns and inflation based on Fama's explanation. His results show a positive relation between real output production and inflation as predicted by the Phillips curve (Kryzanowski and Rahman, 2009). Their findings are in contrast to the premise of the proxy hypothesis.

    Geske and Roll (1983) offer an alternative hypothesis. Their interpretation is that a decline in stock prices indicates a slowdown in economic activity, which puts a strain on government revenue and heralds budget deficits. The anticipated government borrowing to fund debts may be financed by monetization. As a result, a rise in money supply triggers an upward shift in inflation expectations. This process explains the negative relation between stock market returns and expected inflation. Solnik (1983) examines the data for nine countries and fails to find a significant connection between real returns and inflationary expectations. This finding results in a rejection of the relation between stock returns and inflationary expectations for each major stock market. Using a VARMA model to examine the relation among stock returns, real activity, inflation and money supply change, James et al. (1985) find evidence in line with Geske and Roll's hypothesis. Kaul (1987) provides a different explanation for the relation between stock market returns and expected inflation. It can be argued that a decline in the stock market signals weaker economic activities, which can induce a central bank to generate a quantity easing, that in turn, gives rise to a higher inflation expectation. Due to the fact that the U.S. mainly adopted a counter-cyclical approach to its monetary policy for most of the post-war era, Kaul (1987) finds evidence to support a negative correlation between stock market returns and expected inflation using this time period.

    The causality of economic variables is rather complicated depending on the sources of shocks or initial condition. Geske and Roll's (1983) model starts with a decline in real activity that causes government deficits, which are financed by monetization and in turn result in inflation. This scenario is inconsistent with the current market condition that we face. In order to stomp out inflation, the Fed adopted an aggressive monetary policy stance with multiple rate hikes, causing significant stock market volatility that led the stock market to plunge. This inflation-stock volatility-stock return channel has not been comprehensively studied although the issue was mentioned in the literature. For instance, Saryal (2007) examines the data in two countries, Turkey and Canada, with vastly different rates of inflation. Her study finds that the rate of inflation is highly predictive of stock market volatility in Turkey, whereas it has a weaker but still significant impact on the Canadian market. Chiang (2023) also incorporates the equity market volatility into the real stock return-inflation specification and finds a significant negative relation between real stock index returns and expected inflation for 12 major economies. Recently, Batten et al. (2021) finds that during the GFC, investors reduced their stock positions and conclude that implied volatility shocks adversely affect the portfolio returns of stock oil hedges. In a follow up study, Batten et al. (2023) report evidence that VIX has a significant impact on European bank returns. Moreover, in a study using the bibliometric analysis to examine the stock market returns-inflation relation, Hoong et al. (2023) argue that it is vital to integrate theoretical implications in light of the volatile market conditions and rising inflation rate in the empirical analysis.

    Despite the significant insight that can be gained from using market volatility to explain the relation between real stock market returns and inflation expectations, very little research has explored this market volatility channel and the potential resulting relations. Our aim of this paper is to present evidence to validate the links between (real) stock market returns and inflation expectations via an EMV channel. These linkages can be described by the following two paths. First, heightened inflation in the U.S. leads to a rise in EMV (Saryal, 2007; Baker et al., 2022; Chiang, 2023; Chiang and Tang, 2023); second, heightened market volatility in the U.S. causes decline stock returns in the U.S., which has a spillover effect to global markets through financial contagion.2

    2There are studies focusing on sectoral investigation. For instance, Bouri et al. (2023) find energy and materials have the capacity to hedge against inflation risk. Kilian and Zhou (2021) find a significant short-run impact of rising oil prices on U.S. inflation.

    The above discussions lead us to specify a regression model expressed as:

    Rt=C+β1πt+β2ICrisis+β3ICOVID19+εt (1)

    where β1<0, β2<0, β3<0, The error term, εtΩt1GED(0,σ2t1,ν).

    Equation (1) is a mean equation expressing the stock market return, Rt, is related to inflation, πt (or expected inflation, πet) (Fama, 1981) while controlling for the extreme observations captured by indicator variables, ICrisisandICOVID19 for the 2008–2009 global financial crisis (GFC) and 2019–2020 COVID-19 pandemic (Cheema et al., 2020; Terry et al., 2002). The variance equation follows the specification of Ding et al. (1993) and features an asymmetric power GARCH (APARCH) model, which is given by:

    σδt=ω+pi=1αi(|εtj|γiεti)δ+qi=1βjσδtj (2)

    The model contains a power transformation (δ) of the conditional standard deviation process, which deviates from absolute residuals; where δ > 0, |γi|1 for i = 1, 2, ..., r; γi = 0 for all i > r, and rp If the null hypothesis γi = 0 for all i is rejected, the result suggests an asymmetric effect. It should be noted that ARCH, GARCH and GJR-GARCH models are all nested in this specification.

    To simplify the process, it is convenient to impose j = i = 1 in (2). The specification for APARCH (1, 1) model can be expressed as:

    σδt=ω+α(|εt1|γεt1)δ+βσδt1. (3)

    The parameters to be estimated involve calculations of ω, α,γ, β and δ. The specification of (3) is appealing since it is simple yet captures different properties of heteroscedasticity of variance and allows the data to determine an optimal δ value.

    We follow Nelson (1991) and Li et al. (2005), who assume that εt obeys GED distribution, which is expressed as:

    εtΦt1GED(0,σ2t1,υ) (4)

    The use of GED distribution allows us to deal with leptokurtotic property (fat tails), which is often embodied in asset series as noted by Li et al. (2005) and Chiang (2020).

    The data in this study cover the sample period from December 1989 through June 2022 (and extended to May 2023 for robustness test) for 20 countries; some countries such as China and Indonesia have later started points for the data. The data contain aggregate stock indices, including the United States (US), Canada (CA) United Kingdom (UK), France (FR), Germany (GM), Italy (IT), Japan (JP), Brazil (BR), Russia (RU), India (IN), China (CN), South Africa (SA), Spain (ES), the Netherlands (NE), Sweden (SD), Switzerland (SW), Turkey (TK)), Indonesia (ID), South Korea (KO) and Mexico (MX). The same countries are selected for consumer price index (CPI). The selection of these countries is built around the G20 or top 20 GDP. Argentina, Australia and Iran are not included among these countries due to a lack of monthly CPI data available in the source of data base of the Federal Reserve Bank of St. Louis. The stock return or inflation is derived by taking the natural log-difference of the price index times 100. The monthly inflation rates in Table 1 range from 0.04% (Japan) to 4.29% (Brazil), whereas the rate for the U.S. is 0.21%, which is close to the 0.22% for both the U.K. and Italy.

    Table 1.  Summary of monthly inflation rates for 20 countries: January 1990–May 2022.
    πUSt πCAt πUKt πFRt πGMt πITt πJPt πBRt πRUt πINt
    Mean 0.21 0.19 0.22 0.14 0.16 0.22 0.04 4.29 2.65 0.58
    Median 0.19 0.18 0.26 0.13 0.12 0.20 0.00 0.56 0.80 0.62
    Maximum 0.85 2.59 2.42 1.41 2.46 1.58 2.07 60.10 32.52 4.47
    Minimum −0.43 −1.04 −0.70 −1.01 −1.04 −0.68 −1.02 −0.51 −0.54 −2.12
    Std. Dev. 0.13 0.37 0.36 0.29 0.37 0.25 0.36 9.58 5.22 0.81
    Skewness 1.07 0.64 0.96 −0.05 0.74 0.45 1.15 2.86 3.18 0.31
    Kurtosis 7.46 7.38 9.14 4.40 7.52 5.86 8.21 11.48 13.38 5.02
    JB 409 348 691 33 377 150 539 1747 2236 75
    Obs 401 401 401 401 401 401 399 401 362 401
    πCNt πSAt πESt πNEt πSDt πSWt πTKt πIDt πKOt πMXt
    Mean 0.29 0.54 0.24 0.18 0.18 0.09 0.14 0.66 0.29 0.74
    Median 0.19 0.48 0.25 0.15 0.15 0.08 0.00 0.44 0.26 0.55
    Maximum 3.97 2.67 2.98 3.57 2.81 1.19 9.63 11.91 2.50 7.67
    Minimum −1.41 −1.14 −1.93 −1.06 −1.45 −1.05 −3.87 −1.07 −0.75 −1.02
    Std. Dev. 0.85 0.50 0.54 0.48 0.51 0.34 1.23 1.16 0.45 0.86
    Skewness 1.02 0.54 −0.06 0.95 1.29 0.25 3.12 4.95 0.98 2.95
    Kurtosis 5.28 3.97 5.80 8.28 9.54 3.99 23.18 37.32 6.18 17.89
    JB 138 35 131 522 825 21 7472 21376 234 4285
    Obs 352 401 401 398 401 402 402 402 402 401

     | Show Table
    DownLoad: CSV

    The monthly stock market returns in Table 2 range from −0.06% (Japan) to 5.88% (Brazil). Evidence shows that Brazil has a higher stock return along with high inflation. The standard deviations of stock returns indicate that the U.S. (3.66) has the lowest risk compared with Brazil (25.94), which has the highest risk and is about 7 times higher than that of U.S. The Jarque-Bera statistics (JB) in stock market returns for examining normality are significantly large relative to the critical point (5.991) at the 5% level with 2 degrees of freedom, which supports a rejection of the null of normality for real stock return series. The data confirm the relevancy of using GED-APARCH procedure.

    Table 2.  Summary of monthly stock returns for 20 countries: January 1990–June 2022.
    RUSt RCAt RUKt RFRt RGMt RITt RJPt RBRt RRUt RINt
    Mean 0.54 0.41 0.33 0.34 0.33 0.19 −0.06 5.88 1.03 1.10
    Median 0.92 1.05 0.65 1.05 1.02 0.61 −0.10 2.59 1.57 1.51
    Maximum 11.93 11.19 9.99 11.00 13.04 16.80 13.36 418.00 46.38 34.70
    Minimum −25.47 −25.00 −24.19 −28.2 −28.16 −30.69 −24.79 −81.94 −47.89 −27.5
    Std. Dev. 3.66 3.96 3.66 4.61 4.80 5.44 4.76 25.94 9.45 6.60
    Skewness −2.03 −1.87 −1.46 −1.32 −1.37 −0.77 −0.54 10.23 −0.55 −0.06
    Kurtosis 14.87 12.71 10.06 7.84 7.95 6.55 5.02 160.41 10.00 5.86
    JB 2636 1814 978 510 536 251 87 42203 624 137
    Obs 402 402 402 402 402 402 402 402 298 402
    RCNt RSAt RESt RNEt RSDt RSWt RTKt RIDt RKOt RMXt
    Mean 0.38 0.81 0.26 0.42 0.63 0.50 2.74 0.75 0.24 1.35
    Median 0.09 1.49 0.53 1.17 1.31 1.10 2.12 1.34 0.40 1.83
    Maximum 24.67 12.35 13.70 12.65 14.71 14.21 54.36 25.91 20.72 18.49
    Minimum −21.76 −22.18 −29.92 −30.4 −23.90 −18.07 −35.25 −29.48 −23.58 −23.1
    Std. Dev. 6.18 4.37 5.04 4.52 5.04 3.96 10.92 6.43 6.12 5.98
    Skewness 0.26 −1.20 −0.84 −1.76 −0.84 −0.96 0.53 −0.65 −0.16 −0.37
    Kurtosis 4.83 6.93 6.56 11.18 5.83 6.13 5.80 6.66 4.73 4.23
    JB 42 355 259 1328 181 226 150 189 52 34
    Obs 282 402 402 402 402 402 402 300 402 402
    Notes:πit in Table 1 is the monthly inflation rate for country i; Rit is the monthly stock return for country i.

     | Show Table
    DownLoad: CSV

    We also utilize the variable of EMV calibrated to inflation or the interest rate as proposed by Baker et al. (2022), who construct a newspaper-based Equity Market Volatility (EMV) tracker that moves with the CBOE Volatility Index (VIX) (see Table 3) and with realized S & P 500 volatility of returns. EMV refers to three-words that contains: E: {economic}, M: {stock market, equity}, V: {volatility} and their variants, and Π:{cpi, inflation, and the likes}. To construct an Inflation EMV tracker, EMVπ,t, Baker et al. (2022) use the following formula:

    EMVπ,t=(#{EMVΠ}t#{EMV}t)EMVt (5)

    where # denotes the count of newspaper articles in the indicated set, and EMVt is the value of the overall EMV tracker in month t. By using the same procedure, Baker et al. (2022) derive EMVi,t.3

    Table 3.  Correlations among EMV-INF, EMV-INT and VIX for period: 1990M01 2022M08.
    Correlation
    t-Statistic EMV_INF EMV_INT VIX
    EMV_INF 1
    -----
    EMV_INT 0.833 1
    29.74 -----
    VIX 0.415 0.341 1
    9.02 7.16 -----
    Notes: Observations: 392.

     | Show Table
    DownLoad: CSV

    3By the same token, Baker et al. (2022) calculate the share of EMV articles in the category of interest rates and multiply it by the EMV tracker value to obtain category-specific trackers. The measure changes in interest rates that affect EMV during month t is given by EMVi,t, which is simply used to replace EMVπ,t in equation (5). The terms for Interest Rates: {interest rates, yield curve, fed funds rate, overnight rate, repo rate, T-bill rate, bond rate, bond yield} and Monetary Policy: {monetary policy, money supply, open market operations, fed funds rate, discount window, quantitative easing, forward guidance, interest on reserves, Fed chair, the fed}. See (Baker et al., 2022) and the link Policy News and Equity Market Volatility for the details.

    A demonstration of the time series paths of different volatility measures can be seen in Figure 1 that reveals both EMVπ,t and EMVi,t are positively correlated with VIXt. Thus, both EMVπ,t and EMVi,t can be viewed as variants of U.S.VIXt. Moreover, an estimate obtained by regressing emvπ,t (lnEMVπ,t) and emvi,t (lnEMVi,t), respectively, on inflation (πt) shows a positive relation, which is statistically significant. The evidence is reported in Table 4, where emvπ,t and emvi,t are positively correlated with inflation and are statistically significant.

    Figure 1.  Time series plots of VIX and EMVs in relation to inflation and interest rate changes.
    Table 4.  Estimates of equity market volatilities' response to inflation in U.S. market using a GARCH(1, 1) model.
    Dept. variable C πt IGFC,t ICOVID,t ω ε2t1 σ2t1 ¯R2
    emvπ,t 1.537 0.541 0.416 0.510 0.086 0.334 0.817 0.06
    178.37 17.07 3.10 9.29 0.37 0.41 1.94
    emvi,t 1.639 0.373 0.127 0.184 0.021 0.196 0.942 0.01
    777.23 62.03 0.98 47.48 0.58 0.50 10.86
    Notes: emvπ,t= ln (EMVπ,t) denotes natural log of equity market volatility calibrated to inflation, emvi,t=ln(EMVi,t) denotes natural log of equity market volatility in relation to interest rate change, πt is inflation in U.S., IGFC,t is the indicator variable for 2008 global financial crisis, ICOVID,t is for the COVID-19 in 2020.

     | Show Table
    DownLoad: CSV

    Estimates of Equations (1) with (3) and (4) are shown in Table 5. Several empirical findings merit comments. First, the correlations between stock market returns and inflation are mainly negative and statistically significant at the conventional level. These results are consistent with Fama's "proxy hypothesis" (1981). Exceptions are the estimates for Brazil and Russia, which display higher inflation than the other countries under investigation. Yet, the positive sign results for these two countries are consistent with the findings by Solnik and Solnik (1997) and Hasan (2008).

    Table 5.  Estimates of stock return equation by using inflation based on the GED-APARCH(1, 1) process.
    Returns C lnσ2t πt IGFC,t ICOVID,t ω |εt1| εt1 σδt1 δ ¯R2
    RUSt −6.396 2.578 −2.223 −9.870 −1.931 9.711 0.061 −0.458 0.537 2.099 0.12
    −1.66 1.98 −8.78 −8.20 −8.94 1.68 1.42 −3.05 5.32 6.31
    RCAt −11.755 4.093 −0.231 −8.455 −0.078 11.908 0.059 −0.186 0.552 2.090 0.01
    −6.64 6.77 −3.42 −11.46 −0.39 3.29 3.63 −7.08 9.15 11.74
    RUKt −10.006 3.646 −0.285 −5.186 −3.336 10.871 0.087 −0.322 0.513 2.144 0.06
    −12.20 14.16 −3.08 −5.12 −5.94 2.66 2.70 −3.35 3.60 8.94
    RFRt 4.340 −1.163 −1.099 −11.829 −11.044 0.874 0.206 0.585 0.327 0.350 0.05
    3.29 −2.62 −2.64 −9.20 −17.63 4.54 3.33 4.20 4.50 1.76
    RGMt −1.242 0.763 −0.406 −8.117 −1.621 10.054 0.346 −0.007 0.511 2.101 0.01
    −0.89 1.80 −2.79 −5.91 −1.17 1.03 1.63 −0.03 2.68 3.81
    RITt −8.223 2.440 −0.915 −8.589 −2.296 25.284 0.003 −0.510 0.561 2.046 0.01
    −2.00 2.33 −4.05 −7.23 −1.49 1.52 0.17 −0.12 1.19 8.78
    RJPt −7.871 2.027 −0.409 −7.553 −2.447 18.346 0.170 −0.576 0.537 2.083 0.05
    −1.20 1.16 −2.93 −7.91 −2.10 2.70 0.94 −1.80 4.22 9.40
    RBRt 1.291 −3.175 0.874 −7.652 −3.224 0.508 −0.006 −0.561 0.989 1.761 0.14
    10.14 −6.55 79.53 −6.38 −1.75 0.10 −0.08 −0.05 21.80 0.48
    RRUt 6.235 −1.211 0.517 −7.450 −7.934 83.816 0.172 0.045 0.599 2.001 0.02
    0.58 −0.62 3.12 −2.21 −5.18 0.26 0.44 0.81 0.99 1.82
    RINt −3.497 1.422 −0.507 −7.637 −4.111 31.751 0.331 −0.824 0.531 2.088 0.04
    −1.52 2.75 −4.53 −3.98 −2.09 2.69 1.09 −1.43 3.45 17.18
    RCNt −4.236 1.256 −0.898 −6.257 −3.666 29.881 0.170 −0.843 0.541 2.081 0.03
    −1.55 1.97 −6.14 −4.36 −4.01 1.22 0.33 −0.34 2.41 4.44
    RSAt −3.497 1.422 −0.507 −7.637 −4.111 31.751 0.331 −0.824 0.531 2.088 0.04
    −1.52 2.75 −4.53 −3.98 −2.09 2.69 1.09 −1.43 3.45 17.18
    RESt 2.835 −9.146 −0.434 −10.725 −1.891 21.316 0.009 −0.485 0.549 2.058 0.03
    0.31 −0.27 −2.01 −11.29 −1.12 0.32 0.39 −0.47 0.37 2.67
    RNEt −0.733 0.664 −0.583 −7.326 −2.499 14.112 0.197 0.105 0.666 2.644 0.03
    −1.35 3.74 −3.03 −3.83 −1.80 0.36 1.24 0.40 4.04 1.78
    RSDt 3.859 −0.758 −1.013 −4.725 −4.925 18.235 0.463 0.418 0.389 2.175 0.06
    4.95 −3.19 −4.49 −3.63 −2.82 2.24 1.92 2.09 3.15 7.18
    RSWt 0.595 0.195 −0.479 −7.321 −4.470 12.183 0.893 0.198 0.128 1.849 0.02
    1.85 1.94 −4.82 −6.52 −3.42 1.08 2.07 0.97 0.75 4.03
    RTKt −3.738 1.273 −0.468 −13.706 −1.275 52.913 0.384 −0.676 0.667 2.431 0.03
    −3.76 6.01 −3.49 −6.92 −1.48 1.67 2.23 −3.99 12.62 8.42
    RIDt −1.131 0.634 −0.125 −12.839 −5.198 38.763 0.256 0.043 0.603 1.997 0.03
    −0.27 0.72 −7.23 −6.17 −3.04 0.31 1.07 8.53 1.33 1.67
    RKOt −2.902 0.941 −0.440 −5.531 −5.158 18.152 1.063 −0.169 0.594 2.343 0.02
    −2.34 3.18 −2.50 −2.79 −8.10 0.52 1.69 −1.52 4.58 2.48
    RMXt −11.30 3.396 −0.383 −7.209 −0.578 18.909 0.363 −0.22 0.573 2.214 0.06
    −2.76 3.42 −2.46 −6.54 −1.41 2.22 3.45 −3.99 9.33 9.58
    Notes: The variance equation for the US market is expressed as: σ2.099t = 9.711 + 0.061 (|εt1| + 0.458εt1)2.099 + 0.537σ2.099t1. The numbers in the first row are the estimated coefficients, the second row contains the z-statistics. The critical values of z-distribution at the 1%, 5% and 10% levels of significance are 2.58, 1.96, and 1.65, respectively.

     | Show Table
    DownLoad: CSV

    Second, the evidence indicates that the indicator variables for both the 2008–2009 GFC and 2020–2021 COVID-19 present negative signs and are statistically significant. The testing results are consistent with the findings provided by Cheema et al. (2020) and Chiang (2022). Note that these indicator variables were excluded in the previous tests of stock returns-inflation equation. The inclusion of these extreme observations is necessary since those influential observations can produce biased estimators (Tsay, 1988).

    Third, evidence shows that most coefficients of εt1 are negative and significant. These results are consistent with the leverage effect (Glosten et al., 1993). In addition to the coefficient of σδt1 which is positive and whose values of δ are not exactly equal to 2, indicating that the return volatility series presents a clustering phenomenon and an evolution of variance with a long memory (Chen et al., 2018). Taken together, the evidence supports the specification of using an APARCH model.

    Researchers (Kaul, 1990; Chiang, 2023) propose expected inflation can be an alternative variable, which can be used in the test equation. Numerous methods were used to form this expectation. In this study, we extend the adaptive expectation process (Geske and Roll, 1983; Beladi et al., 1993; Kolluri and Wahab, 2008) by adding a seasonal factor. This process is in the spirit of partial adjustment, which is a time series model in the form of an ARIMA(0, 1, 1) process. It is expressed by adding a seasonal factor yield:

    πet=πt1+α(πt1πet1)+γπt12 (6)

    where πt is actual inflation; πet is expected inflation at time t; (πt1πet1) is the forecasting error of inflation; α and γ are parameters following the restrictions of 1≥ α ≥ 0 and 1≥ γ ≥ 0, respectively; πt12 is a seasonal factor. Estimates of expected inflation are reported in Table 6. All the estimated parameters are consistent with the restrictions and satisfy the range of 1 > α > 0. Among the 20 countries, five are absent of seasonal patterns.

    Table 6.  Estimates of expected inflation using an adaptive expectation.
    Expected inflation α γ RMSE
    πet(US) 0.560 - 0.112
    πet(CA) 0.182 0.266 0.345
    πet(UK) 0.281 0.818 0.249
    πet(FR) 0.100 0.564 0.218
    πet(GM) 0.050 0.495 0.312
    πet(IT) 0.385 0.461 0.184
    πet(JP) 0.199 0.446 0.313
    πet(BR) 0.944 - 3.196
    πet(RU) 0.832 - 2.854
    πet(IN) 0.346 0.395 0.710
    πet(CN) 0.408 0.631 0.835
    πet(SA) 0.425 0.310 0.404
    πet(ES) 0.219 0.650 0.373
    πet(NE) 0.194 0.679 0.315
    πet(SD) 0.130 0.445 0.409
    πet(SW) 0.213 0.619 0.261
    πet(TK) 0.256 - 1.167
    πet(ID) 0.609 - 0.972
    πet(KO) 0.379 0.422 0.388
    πet(MX) 0.831 0.428 0.508
    Notes: The expected inflation is assumed to follow an adaptive process, which is equivalent to an exponential smoothing process plus a seasonal factor. In expression, πet = πt1 + α(πt1-πet1 ) + γ. πt12,πt and πet are actual and expected inflation.

     | Show Table
    DownLoad: CSV

    Estimates of adaptive expectation (πet) using Equation (6), as reported in Table 7, show that the coefficients of πet are negative and statistically significant, indicating that stock market returns are negatively related to expected inflation. The evidence is in line with the findings by Solnik (1983) and James et al. (1985). The exceptions again are Brazil and Russia where the coefficients exhibit positive signs. The findings appear to be consistent with the evidence that these two countries present higher than average inflation rates as shown in Table 2. The evidence of positive signs is consistent with the findings by Saryal (2007), Choudhry (2001) and Kumari (2011) in their studies of high inflation countries. The evidence is also in line with Chaudhary and Marrow (2022), who measure investors' expectations using traded inflation-indexed contracts and find that in the post-2000 period, stocks provide positive returns in response to higher expected inflation.

    Table 7.  Estimates of stock returns using adaptive expected inflation based on GED-APARCH(1, 1) process.
    Returns C lnσ2t πet IGFC,t ICOVID,t ω |εt1| εt1 σδt1 δ ¯R2
    RUSt −6.788 2.913 −3.916 −10.134 −2.346 9.317 0.083 −0.207 0.555 2.121 0.13
    −2.03 2.85 −8.72 −7.75 −2.48 4.20 5.48 −3.96 6.09 27.33
    RCAt −7.174 2.669 −0.679 −8.551 −0.649 11.364 0.108 −0.137 0.540 2.132 0.03
    −4.31 4.84 −3.01 −6.43 −0.55 0.66 1.68 −0.92 2.65 2.37
    RUKt −8.303 3.160 −1.360 −4.966 −3.306 10.456 0.113 −0.400 0.535 2.112 0.04
    −1.69 2.02 −10.07 −5.00 −7.30 1.08 2.31 −4.35 5.91 4.13
    RFRt 2.413 −0.543 −0.662 −10.759 −9.727 0.870 0.195 0.586 0.419 0.480 0.06
    1.65 −1.06 −2.00 −8.46 −11.31 1.73 2.72 3.45 3.13 1.32
    RGMt −1.015 0.536 −0.138 −6.334 −0.281 18.836 0.532 0.182 0.530 2.079 0.01
    −1.38 2.84 −1.86 −15.90 −0.20 0.41 0.98 0.63 2.31 1.91
    RITt −8.890 2.653 −1.805 −7.967 −0.033 23.549 0.105 −0.592 0.542 2.074 0.02
    −1.80 2.10 −3.80 −5.11 −0.03 1.15 2.08 −2.40 2.19 6.83
    RJPt −7.871 2.027 −0.409 −7.553 −2.447 18.346 0.170 −0.576 0.537 2.083 0.05
    −1.20 1.16 −2.93 −7.91 −2.10 2.70 0.94 −1.80 4.22 9.40
    RBRt 0.544 −6.608 1.095 −7.235 −4.816 2.871 0.004 1.000 0.948 2.546 0.04
    1.45 −5.53 5.84 −3.96 −2.35 0.38 0.00 0.01 43.10 2.07
    RRUt −0.775 0.509 0.511 −11.101 −13.71 0.189 0.239 0.027 0.752 0.593 0.03
    −0.57 1.17 1.51 −3.96 −12.24 0.89 4.11 0.20 14.50 1.19
    RINt −3.701 1.449 −0.726 −13.748 −5.884 32.29 0.202 −0.443 0.546 2.087 0.04
    −2.04 3.33 −4.29 −9.94 −3.25 0.76 1.18 −4.27 2.81 2.91
    RCNt −4.203 1.198 −0.275 −10.098 −2.910 25.340 0.419 −0.434 0.590 2.147 0.02
    −1.61 2.01 −3.22 −4.60 −5.72 5.73 1.08 −2.68 3.90 11.27
    RSAt −1.371 1.005 −0.975 −5.064 −1.271 13.242 0.212 0.200 0.517 2.123 0.02
    −0.60 1.47 −4.92 −13.52 −0.98 0.25 0.87 0.42 1.65 0.99
    RESt −10.581 3.228 −1.298 −9.780 −1.319 20.595 0.009 −0.689 0.533 2.082 0.03
    −2.24 2.47 −5.94 −7.59 −2.98 0.08 0.54 −0.29 0.86 0.32
    RNEt −0.668 0.637 −0.634 −6.843 −5.488 10.536 0.260 0.042 0.579 2.585 0.05
    −0.65 1.81 −2.54 −2.87 −4.15 0.41 1.48 0.22 3.22 1.76
    RSDt 3.063 −0.588 −0.446 −5.726 −3.426 16.205 0.373 0.311 0.484 2.118 0.05
    1.87 −1.28 −2.26 −3.34 −6.15 0.50 1.03 0.83 2.08 1.93
    RSWt 0.568 0.151 −0.611 −6.419 −4.538 12.921 0.276 0.828 0.261 1.855 0.03
    1.03 0.94 −6.98 −5.64 −3.09 3.08 0.40 0.35 1.30 7.24
    RTKt −3.539 1.199 −0.500 −13.612 −3.496 85.010 0.569 −0.333 0.458 2.200 0.02
    −1.08 1.87 −6.10 −6.27 −1.80 1.01 1.93 −2.27 2.52 6.11
    RIDt −2.107 0.998 −0.871 −7.282 −7.076 35.909 0.476 0.833 0.602 2.020 0.01
    −18.41 39.56 −17.57 −3.43 −5.54 7.25 1.36 1.47 13.88 17.83
    RKOt −3.344 1.096 −0.707 −7.809 −2.359 26.872 0.607 −0.056 0.514 2.180 0.01
    −8.55 10.63 −2.75 −20.73 −11.93 1.16 3.10 −0.37 2.81 4.33
    RMXt −11.32 3.367 −0.181 −7.629 −2.160 26.739 0.207 −0.236 0.547 2.084 0.03
    −8.38 13.76 −3.06 −9.11 −1.62 1.93 2.76 −2.63 4.09 9.54
    Notes: The variance equation for the U.S. market is: σ2.121t = 9.317+ 0.083 (|εt1| + 0.207εt1)2.121 + 0.555σ2.121t1. The numbers in the first row are the estimated coefficients, the second row contains the z-statistics. The critical values of z-distribution at the 1%, 5% and 10% levels of significance are 2.58, 1.96, and 1.65, respectively.¯R2 is the adjusted R-squared.

     | Show Table
    DownLoad: CSV

    It is useful to conduct a robustness test using a different measure of inflation expectations. To this end, the lagged inflation was used as an instrument to predict inflation as denoted by πeIV,t. This is equivalent to imposing the restriction of α = γ = 0 in Equation (6). The estimated results, which are reported in Table 8, show that the estimated coefficients and the related statistical results are comparable. These results include the positive coefficients for Brazil and Russia, two countries that have higher inflation rates and comparable higher stock returns; these market forces drive stock returns and expected inflation, which are positively correlated. With respect to the other variables, statistical performances are maintained as demonstrated by similar qualitative results. The evidence shows the model is robust whether the expectations are formed by an adaptive fashion or using the lagged value to serve as an instrumental variable.

    Table 8.  Robust test for stock returns using expected inflation by Ⅳ based on the GED-APARCH(1, 1) process.
    Returns C lnσ2t πeIV,t IGFC,t ICOVID,t ω |εt1| εt1 σδt1 δ ¯R2
    RUSt −6.788 2.913 −3.916 −10.134 −2.346 9.317 0.083 −0.207 0.555 2.121 0.13
    −2.03 2.85 −8.72 −7.75 −2.48 4.20 5.48 −3.96 6.09 27.33
    RCAt −7.174 2.669 −0.679 −8.551 −0.649 11.364 0.108 −0.137 0.540 2.132 0.03
    −4.31 4.84 −3.01 −6.43 −0.55 0.66 1.68 −0.92 2.65 2.37
    RUKt −8.303 3.160 −1.360 −4.966 −3.306 10.456 0.113 −0.400 0.535 2.112 0.04
    −1.69 2.02 −10.07 −5.00 −7.30 1.08 2.31 −4.35 5.91 4.13
    RFRt 2.413 −0.543 −0.662 −10.759 −9.727 0.870 0.195 0.586 0.419 0.480 0.06
    1.65 −1.06 −2.00 −8.46 −11.31 1.73 2.72 3.45 3.13 1.32
    RGMt −1.015 0.536 −0.138 −6.334 −0.281 18.836 0.532 0.182 0.530 2.079 0.01
    −1.38 2.84 −1.86 −15.90 −0.20 0.41 0.98 0.63 2.31 1.91
    RITt −8.890 2.653 −1.805 −7.967 −0.033 23.549 0.105 −0.592 0.542 2.074 0.02
    −1.80 2.10 −3.80 −5.11 −0.03 1.15 2.08 −2.40 2.19 6.83
    RJPt −7.871 2.027 −0.409 −7.553 −2.447 18.346 0.170 −0.576 0.537 2.083 0.03
    −1.20 1.16 −2.93 −7.91 −2.10 2.70 0.94 −1.80 4.22 9.40
    RBRt 0.544 −6.608 1.095 −7.235 −4.816 2.871 0.004 1.000 0.948 2.546 0.04
    1.45 −5.53 5.84 −3.96 −2.35 0.38 0.00 0.01 43.10 2.07
    RRUt −0.775 0.509 0.511 −11.101 −13.71 0.189 0.239 0.027 0.752 0.593 0.03
    −0.57 1.17 1.51 −3.96 −12.24 0.89 4.11 0.20 14.50 1.19
    RINt −3.701 1.449 −0.726 −13.748 −5.884 32.29 0.202 −0.443 0.546 2.087 0.04
    −2.04 3.33 −4.29 −9.94 −3.25 0.76 1.18 −4.27 2.81 2.91
    RCNt −4.203 1.198 −0.275 −10.098 −2.910 25.340 0.419 −0.434 0.590 2.147 0.02
    −1.61 2.01 −3.22 −4.60 −5.72 5.73 1.08 −2.68 3.90 11.27
    RSAt −1.371 1.005 −0.975 −5.064 −1.271 13.242 0.212 0.200 0.517 2.123 0.02
    −0.60 1.47 −4.92 −13.52 −0.98 0.25 0.87 0.42 1.65 0.99
    RESt −10.581 3.228 −1.298 −9.780 −1.319 20.595 0.009 −0.689 0.533 2.082 0.03
    −2.24 2.47 −5.94 −7.59 −2.98 0.08 0.54 −0.29 0.86 0.32
    RNEt −0.668 0.637 −0.634 −6.843 −5.488 10.536 0.260 0.042 0.579 2.585 0.05
    −0.65 1.81 −2.54 −2.87 −4.15 0.41 1.48 0.22 3.22 1.76
    RSDt 3.063 −0.588 −0.446 −5.726 −3.426 16.205 0.373 0.311 0.484 2.118 0.05
    1.87 −1.28 −2.26 −3.34 −6.15 0.50 1.03 0.83 2.08 1.93
    RSWt 0.568 0.151 −0.611 −6.419 −4.538 12.921 0.276 0.828 0.261 1.855 0.03
    1.03 0.94 −6.98 −5.64 −3.09 3.08 0.40 0.35 1.30 7.24
    RTKt −3.539 1.199 −0.500 −13.612 −3.496 85.010 0.569 −0.333 0.458 2.200 0.02
    −1.08 1.87 −6.10 −6.27 −1.80 1.01 1.93 −2.27 2.52 6.11
    RIDt −2.107 0.998 −0.871 −7.282 −7.076 35.909 0.476 0.833 0.602 2.020 0.01
    −18.41 39.56 −17.57 −3.43 −5.54 7.25 1.36 1.47 13.88 17.83
    RKOt −3.344 1.096 −0.707 −7.809 −2.359 26.872 0.607 −0.056 0.514 2.180 0.01
    −8.55 10.63 −2.75 −20.73 −11.93 1.16 3.10 −0.37 2.81 4.33
    RMXt −11.32 3.367 −0.181 −7.629 −2.160 26.739 0.207 −0.236 0.547 2.084 0.03
    −8.38 13.76 −3.06 −9.11 −1.62 1.93 2.76 −2.63 4.09 9.54
    Notes: The variance equation is: σδt=ω+pi=1αi(|εt1|γiεti)δ+βσδt1. For the U.S. market, it is expressed as: σ2.121t = 9.317+ 0.083 (|εt1| + 0.207εt1)2.121 + 0.533σ2.121t1. The numbers in the first row are the estimated coefficients, the second row contains the z-statistics. The critical values of z-distribution at the 1%, 5% and 10% levels of significance are 2.58, 1.96, and 1.65, respectively.¯R2 is the adjusted R-squared.

     | Show Table
    DownLoad: CSV

    This section introduces a new variable, ΔEMVπ,t, which is the change in EMV calibrated to U.S. inflation. This variable reflects the change of forward-looking stock market volatility as reported in economic news (Terry et al., 2020; Baker et al., 2022). Incorporating the argument of Δemvπ,t into the model, the mean equation becomes:

    Rt=C+β1lnˆσ2t+β2πet+β3Δemvπ,t+β4ICrisis+β5ICOVID19+εt (7)

    where Δemvπ,t is a change in EMV calibrated to the U.S. inflation rate. The lower case emvπ,t denotes ln (EMVπ,t). The restriction β3<0 suggests that as emvπ,t intensifies, investors' fears will increase and lead to a selloff that causes stocks to plunge. Note that the term of EMVπ,t differs from the risk term of Cieslak and Pflueger (2023), who specify an inflation risk term as the covariance between consumption and expected inflation in their model.

    Table 9 reports the estimates of stock market returns in response to expected inflation with an addition of a Δemvπ,t term utilizing the GED-APARCH(1, 1) procedure. The explanatory power for the estimated equations, interestingly, is improved as reflected in higher or at least equal adjusted R-squared. In reviewing results for Brazil and Russia, the coefficients of expected inflation for both countries continue to show positive signs, but only Brazil's coefficient is positive and statistically significant. This result is similar to that of Tripathi and Kumar (2015), who find that only Brazil demonstrates a positive relation among the BRICS countries. The evidence shows that Brazil has been experiencing a monthly high inflation rate (4.29%) and high stock returns (5.88%). Importantly, evidence for 18 out of 20 countries is negative and statistically significant, supporting the inflation-stock return relation as proposed by Fama (1981) or Geske and Roll (1983).

    Table 9.  Estimates of stock returns in relation to conditional variance, expected inflation (Ⅳ approach) and change in EMV controlling for global financial crisis and COVID-19 using the APARCH(1, 1) model.
    Returns C lnσ2t πeIV,t Δemvπ,t IGFC,t ICOVID,t ω |εt1| εt1 σδt1 δ ¯R2
    RUSt 0.159 0.499 −1.885 −0.022 −8.870 −1.325 0.143 0.106 0.276 0.858 0.738 0.19
    0.28 1.90 −1.99 −6.28 −7.70 −1.04 1.33 3.32 0.99 18.26 1.77
    RCAt −1.237 0.759 −0.658 −0.032 −6.566 0.369 13.19 0.113 −0.085 0.594 2.014 0.08
    −0.81 1.75 −3.53 −16.98 −5.15 1.38 0.41 2.83 −1.02 1.35 1.69
    RUKt −3.683 1.652 −0.415 −0.021 −6.598 −3.347 9.576 0.068 −0.240 0.561 2.061 0.11
    −1.44 2.00 −2.07 −12.25 −6.84 −2.75 1.44 1.98 −2.37 1.41 2.97
    RFRt 3.705 −1.067 −0.040 −0.035 −8.934 −7.754 0.626 0.122 0.546 0.707 0.904 0.15
    1.65 −1.30 −0.06 −6.36 −6.24 −9.33 1.28 2.50 2.34 6.82 1.68
    RGMt 5.146 −1.620 −0.295 −0.038 −10.15 −9.023 2.638 0.36 0.251 0.031 0.904 0.13
    2.86 −2.55 −0.57 −9.84 −12.98 −12.69 1.90 4.01 2.12 0.27 2.44
    RITt −3.27 1.182 1.282 −0.027 −8.452 −2.972 25.09 0.14 −0.092 0.577 2.033 0.05
    −0.56 0.83 −3.01 −11.76 −7.81 −2.03 0.21 0.71 −0.21 1.40 1.71
    RJPt −3.906 1.220 0.200 −0.028 −11.479 −6.944 0.896 0.054 0.441 0.884 1.605 0.03
    −1.87 1.80 −0.31 −4.95 −15.92 −9.44 0.63 2.32 1.56 25.16 1.64
    RBRt −0.222 −0.020 1.893 −0.035 −8.027 −9.747 82.105 0.084 0.998 0.594 3.097 0.04
    −0.51 −0.92 13.31 −3.17 −1.93 −4.69 0.37 0.00 0.00 4.46 2.34
    RRUt 0.553 0.145 0.483 −0.020 −10.92 −13.15 0.416 0.261 −0.011 0.751 1.073 0.04
    0.39 0.31 1.30 −2.39 −3.01 −12.51 0.82 3.68 −0.08 11.77 1.82
    RINt −2.594 1.265 0.626 −0.017 −14.40 −6.715 29.83 0.497 −0.665 0.569 2.121 0.04
    −1.01 2.17 −4.67 −5.87 −6.19 −3.76 1.00 1.93 −2.75 4.20 4.52
    RCNt −4.812 1.320 0.404 −0.013 −7.497 −1.558 5.953 0.290 −0.340 0.779 2.562 0.02
    −3.15 3.31 − 2.01 −5.46 −5.04 −2.78 0.77 2.00 −2.68 19.50 3.10
    RSAt 0.763 0.282 0.475 −0.030 −5.627 −3.297 15.86 0.250 0.097 0.576 2.030 0.08
    0.38 0.52 −2.83 −11.59 −6.10 −1.58 0.30 1.47 1.02 1.39 1.68
    RESt −3.996 1.361 0.981 −0.034 −8.431 −0.959 20.046 0.104 −0.604 0.530 2.084 0.09
    −3.54 4.77 −7.01 −12.46 −5.29 −0.52 1.53 1.17 −1.19 2.06 7.50
    RNEt 2.477 −0.457 0.390 −0.029 −6.131 −2.009 5.473 0.178 0.536 0.555 2.109 0.14
    2.83 −1.42 −1.76 −8.87 −4.29 −1.50 0.68 0.95 0.84 3.27 2.32
    RSDt 4.482 −1.074 0.542 −0.038 −2.110 −1.277 17.529 0.151 0.676 0.565 2.090 0.11
    1.74 −1.59 −4.24 −17.95 −6.76 −0.78 0.43 1.90 1.07 3.36 1.79
    RSWt 1.966 −0.238 0.773 −0.017 −4.761 −4.638 24.937 0.335 0.628 0.200 2.648 0.08
    3.45 −1.18 −3.94 −8.27 −3.51 −3.31 0.54 0.53 0.62 0.80 2.08
    RTKt −3.273 1.187 0.419 −0.009 −11.214 −5.478 83.319 0.822 −0.701 0.578 2.093 0.02
    −1.15 2.34 −7.06 −2.23 −5.05 −3.56 0.30 1.10 −2.27 5.05 1.80
    RIDt −1.693 0.962 0.884 −0.010 −12.623 −6.478 37.013 0.162 0.050 0.600 2.002 0.04
    −0.25 0.67 −5.56 −2.83 −4.85 −3.05 0.45 0.64 1.82 1.21 1.89
    RKOt −2.889 0.942 1.303 −0.033 −4.851 −3.524 21.296 0.975 −0.185 0.467 2.197 0.06
    −5.94 7.64 −7.93 −15.77 −3.36 −3.09 2.19 3.22 −2.13 3.16 6.27
    RMXt −1.007 0.727 0.019 −0.029 −9.105 −5.583 0.016 0.082 −0.015 0.923 2.018 0.10
    −1.41 2.78 −0.42 −5.49 −6.05 −5.65 0.71 1.94 −0.07 34.86 1.96
    Notes: The mean equation is: RUSt=C+β1logˆσ2i,t+β2ˆπet+β3ΔEMVπ,t+β7DCrisis + β8 DCOVID19 + ε,t. The variance equation for the U.S. market is: σ0.738t = 0.143 + 0.106 (|εt1|- 0.276εt1)0.738 + 0.858.σ0.738t1. The numbers in the first row are the estimated coefficients, the second row contains the z-statistics. The critical values of z-distribution at the 1%, 5% and 10% levels of significance are 2.58, 1.96, and 1.65, respectively.¯R2 is the adjusted R-squared.

     | Show Table
    DownLoad: CSV

    A most noticeable result in Table 9 is the estimated coefficients of Δemvπ,t, which present negative signs and are strongly significant for all countries. This outcome is consistent with the phenomenon that as inflation rises in U.S., the input costs of production increases which cuts profits as well as future cash flows; as a result, investors' fears grow, heightening stock market volatility in U.S. market. This heightened volatility further spills over to the rest of world through the contagion effect (Koutmos and Booth, 1995; Karolyi, 1995; Chiang et al., 2007; Chen et al., 2018; Wang et al., 2018; Chiang, 2020). This spillover explains the negative correlations between global stock market returns and U.S. equity market volatility. In addition, the latter is also positively correlated with U.S. inflation as reported in Table 4. Thus, the evidence supports the uncertainty hypothesis, which runs from πUStΔemvπ,tRit. This link differs from Fama's proxy hypothesis, which relies on a connection with real activity (qt), that is, πUStΔqtRit.4 Our testing result is consistent with the studies by Bessler and Yang (2003) and Rapach et al. (2013) as they show that the U.S. stock market plays a significant role in the price movements of global stock returns. Yet, this study provides a new channel for understanding the negative effect that US Δemvπ,t has on global stock returns rather than via a channel of purchasing power parity observed by Yang et al. (2006). Note that the negative coefficient of Δemvπ,t implies that a risk premium will be awarded to investors for taking the excessive risk when they buy stocks at time t-1.5 Yet, this finding differs from Sarte (1998), who finds the inflation risk is not significant due to the fact that inflation surprises and unexpected movements in consumption growth exhibit little covariation in U.S. data.

    4The relation of output growth (Δqt) in relation to inflation (πt) in the U.S. is as follows. The correlation is positive and highly significant. This evidence is in contrast to Fama's assertion (1981).

    Dept. variable C πt IGFC,t ICOVID,t ω |εt1| σδt1 R2
    Note: The first-row numbers are the estimated coefficients, the row below are the z-statistics. Δqt and πt are positively correlated.
    Δqt 0.018 0.587 -0.819 -0.670 0.314 0.707 0.262 0.08
    0.48 3.35 -1.52 -2.74 1.80 2.58 1.15

    5Using U.K. in Table 9 as an example, the coefficient of Δemvπ,t in RUKt=-0.021Δemvπ,t = -0.021emvπ,t + 0.021emvπ,t1, predicts a positive relation between RUKt and emvπ,t1, which is consistent with mean-variance analysis.

    Comparable results are produced for the rest of the other explanatory variables as shown in Table 9. However, the coefficients of risk variable (lnσ2t) become less significant or even change sign due to the dominance of Δemvπ,t. It appears that the Δemvπ,t has more information content as compared with that of lnσ2t due to the fact that the former contains more explanatory power from news sources. It is interesting to compare the test results in Table 9 against the conventional models of Balduzzi (1995) and Li et al. (2010). Specifically, this study appears to have more information content, which is attributable to its improved model specifications related to the expected inflation variable, its ability to capture the risk factor, Δemvπ,t and the power δ of the heteroskedasticity equation.

    In the Fisher apparatus, inflation is viewed as a monetary phenomenon, which implies no real effect on economic activities. However, during highly volatile market conditions, the change in expected inflation can have an impact on real stock market returns since adjustments in the stock market and commodity prices do not move at the same pace as the release of news about inflation. By recognizing this possibility, this section estimates the real stock returns in relation to expected inflation and changes in EMV. The estimates using two different measures of expected inflation are shown in Tables 10 and 11.

    Table 10.  Estimates of real stock returns using conditional variance, expected inflation (adaptive approach) and change in EMV controlling for global financial crisis and COVID-19 using the GED-APARCH(1, 1) model.
    Real Returns C lnσ2t πet Δemvπ,t IGFC,t ICOVID,t ω |εt1| εt1 σδt1 δ ¯R2
    rUSt 0.834 0.138 −2.611 −0.024 −10.32 −5.087 0.133 0.122 0.124 0.844 0.662 0.22
    1.16 0.44 −2.05 −6.80 −9.73 −8.61 1.21 2.87 0.62 15.13 1.29
    rCAt −1.964 1.001 −1.851 −0.030 −5.339 0.320 11.750 0.155 −0.112 0.567 2.075 0.10
    −2.16 3.76 −6.74 −14.71 −3.40 2.09 1.12 2.01 −2.85 2.07 2.86
    rUKt −3.714 1.660 −2.299 −0.025 −6.489 −3.653 8.695 0.119 −0.005 0.538 2.108 0.16
    −5.01 6.41 −11.24 −12.42 −9.71 −3.62 0.80 1.83 −0.05 2.94 2.84
    rFRt 4.586 −1.377 −2.608 −0.038 −9.329 −8.265 0.550 0.118 0.669 0.718 0.712 0.15
    1.94 −1.63 −3.54 −6.60 −6.57 −11.02 1.50 3.04 3.58 7.85 1.68
    rGMt 15.495 −4.907 −0.936 −0.053 −9.494 −9.026 5.778 0.447 0.259 −0.150 1.185 0.01
    7.84 −8.13 −2.06 −19.43 −12.65 −27.75 2.45 5.37 5.08 −3.33 4.43
    rITt −3.088 1.091 −1.635 −0.028 −8.160 −0.192 25.599 0.155 0.009 0.585 2.022 0.03
    −2.11 3.32 −2.91 −11.45 −11.60 −2.47 2.72 1.03 0.12 1.67 5.37
    rJPt 3.562 −1.16 −1.177 −0.021 −6.275 −2.747 0.124 0.040 0.786 0.907 0.442 0.08
    1.50 −1.45 −1.52 −4.23 −4.10 −1.37 1.93 3.12 3.90 66.78 1.15
    rBRt 106.39 −21.27 0.222 −0.018 −7.440 −4.266 297.073 −0.001 0.996 0.327 2.453 0.02
    2.08 −2.06 3.38 −1.12 −1.65 −0.86 0.29 −0.03 0.03 2.32 1.86
    rRUt −0.258 0.212 0.303 −0.021 −10.576 −4.085 0.353 0.222 0.135 0.772 0.899 0.04
    −0.16 0.40 0.60 −2.51 −2.79 −3.86 0.83 2.99 0.72 12.03 1.59
    rINt −2.057 0.918 −1.106 −0.011 −13.705 −5.629 38.219 0.471 −0.732 0.569 2.040 0.05
    −4.41 8.66 −9.97 −4.39 −9.94 −3.01 1.90 3.02 −3.62 3.86 6.80
    rCNt −3.163 0.899 −0.786 −0.007 −6.455 −3.454 33.908 0.799 −0.860 0.587 2.043 0.03
    −2.21 3.28 −4.99 −2.04 −3.15 −5.69 0.60 3.06 −2.08 5.09 3.00
    rSAt 1.616 0.008 −2.005 −0.021 −5.145 −2.726 10.570 0.186 0.581 0.582 2.115 0.12
    18.10 0.99 −10.61 −12.30 −6.55 −1.37 0.21 0.41 0.36 1.50 0.83
    rESt −4.695 1.543 −1.620 −0.031 −10.84 0.389 19.352 0.217 −0.444 0.531 2.089 0.10
    −0.94 1.17 −10.24 −12.40 −8.34 0.42 0.42 0.91 −1.92 3.19 1.76
    rNEt 2.280 −0.360 −0.832 −0.028 −4.593 −1.744 8.912 0.151 1.000 0.533 2.122 0.13
    3.01 −1.48 −4.70 −10.71 −3.17 −4.86 1.75 0.05 0.05 3.60 4.74
    rSDt 4.173 −0.917 −1.107 −0.037 −2.521 −2.502 16.627 0.244 0.589 0.538 2.118 0.13
    4.93 −3.65 −6.82 −23.80 −1.79 −1.63 2.43 1.20 1.17 3.73 6.19
    rSWt 1.531 −0.141 −2.024 −0.015 −5.075 −2.173 5.498 0.313 0.236 0.436 1.979 0.08
    2.23 −0.55 −7.29 −5.56 −3.86 −1.83 0.44 1.37 0.69 1.65 1.21
    rTKt −2.056 0.997 −0.458 −0.004 −10.83 −2.822 111.016 0.281 −0.371 0.599 2.002 0.01
    −0.27 0.76 −2.17 −1.13 −4.89 −2.61 0.18 1.26 −1.14 1.56 1.13
    rIDt −2.137 0.874 −1.712 −0.002 −4.892 −4.892 25.860 0.830 −0.406 0.528 2.218 0.12
    −1.18 2.07 −10.40 −1.05 −3.63 −2.94 0.60 1.41 −2.53 2.96 2.96
    rKOt −3.660 1.136 −2.464 −0.030 −3.72 −2.935 21.125 0.983 −0.176 0.539 2.161 0.07
    −2.24 3.06 −8.44 −18.46 −5.55 −2.33 2.34 2.76 −1.75 3.93 7.50
    rMXt −1.090 0.733 −0.734 −0.024 −9.80 −5.461 0.013 0.097 −0.066 0.913 0.467 0.08
    −1.50 2.56 −2.32 −4.39 −8.89 −5.30 0.54 2.58 −0.38 36.34 0.85
    Notes: The lower case rt is the real stock return, rit= Ritπit for country i. πet is expected inflation, ΔEMVπ,t is a change in equity market volatility calibrated to inflation. The numbers in the first row are the estimated coefficients, the second row contains the z-statistics. The critical values of z-distribution at the 1%, 5% and 10% levels of significance are 2.58, 1.96, and 1.65, respectively.¯R2 is the adjusted R-squared.

     | Show Table
    DownLoad: CSV
    Table 11.  Estimates of real stock returns in relation to conditional variance, expected inflation, EMV calibrated to interest rate changes controlling for global financial crisis and COVID-19 using the GED-APARCH(1, 1) model.
    Real Returns C lnσ2t πet Δemvi,t IGFC,t ICOVID,t ω |εt1| εt1 σδt1 δ ¯R2
    rUSt 0.585 0.308 −2.845 −0.022 −9.056 −4.344 0.100 0.119 0.162 0.842 0.424 0.21
    1.28 1.48 −2.75 −7.32 −7.92 −7.82 1.30 2.64 0.79 22.31 0.88
    rCAt −2.965 1.294 −1.704 −0.024 −4.947 0.653 10.821 0.154 −0.201 0.549 2.093 0.11
    −1.30 1.87 −7.67 −16.30 −3.03 2.01 2.00 4.96 −3.97 2.54 9.03
    rUKt −2.703 1.341 −2.435 −0.014 −3.955 −3.004 9.528 0.109 −0.177 0.560 2.062 0.09
    −0.63 0.97 −11.33 −6.65 −4.82 −2.60 0.34 0.88 −0.56 1.48 1.19
    rFRt 2.283 −0.646 −2.373 −0.035 −9.181 −7.526 0.449 0.126 0.687 0.702 0.530 0.14
    1.21 −0.93 −3.43 −12.19 −6.26 −11.25 1.78 3.43 4.12 9.93 1.47
    rGMt 0.040 0.332 −1.538 −0.030 −9.123 −7.521 1.150 0.363 0.031 0.221 0.579 0.10
    0.73 4.22 −3.01 −7.50 −8.53 −7.49 1.63 3.15 0.23 1.34 1.47
    rITt −2.775 1.024 −2.226 −0.029 −8.178 −5.829 26.753 0.139 −0.432 0.599 2.001 0.06
    −1.93 3.02 −4.47 −13.65 −11.76 −3.37 0.31 0.85 −0.87 1.47 1.27
    rJPt 0.215 −0.060 −1.244 −0.024 −7.300 −2.571 17.195 0.331 0.040 0.558 2.061 0.07
    0.11 −0.12 −3.66 −11.36 −4.74 −2.50 1.01 1.10 0.53 1.17 3.21
    rBRt −0.867 0.431 0.098 −0.030 −6.583 −5.778 0.761 −0.015 0.088 1.001 2.237 0.01
    −0.25 0.45 1.02 −3.68 −3.09 −2.97 0.83 −2.06 0.31 77.96 3.55
    rRUt 1.367 −0.036 −0.395 − 0.024 −2.264 −5.673 1.151 0.251 −0.012 0.797 1.715 0.01
    1.32 −0.11 −1.24 −4.50 −1.31 −3.66 0.35 1.94 −0.05 8.35 1.23
    rINt −2.114 0.950 −0.974 −0.012 −14.073 −5.389 42.372 0.168 −0.409 0.600 2.001 0.03
    −2.32 5.63 −6.67 −6.49 −10.68 −2.79 1.14 1.41 −2.24 1.88 4.23
    rCNt −3.163 0.899 −0.786 −0.007 −6.455 −3.454 33.908 0.799 −0.860 0.587 2.043 0.03
    −2.21 3.28 −4.99 −2.04 −3.15 −5.69 0.60 3.06 −2.08 5.09 3.00
    rSAt 1.761 −0.054 −1.711 −0.027 −7.146 −2.760 12.210 0.185 0.664 0.547 2.083 0.14
    3.10 −0.33 −11.76 −17.38 −8.52 −1.41 0.46 0.27 0.25 1.59 1.72
    rESt −5.273 1.738 −1.455 −0.032 −11.566 0.392 19.812 0.042 −0.638 0.543 2.065 0.08
    −1.63 1.99 −9.49 −16.61 −10.11 0.51 0.21 0.60 −0.67 1.23 0.86
    rNEt 2.583 −0.355 −1.634 −0.026 −6.773 −1.860 14.348 0.197 0.831 0.533 2.172 0.14
    5.00 −2.41 −13.10 −15.75 −3.62 −2.03 3.79 0.35 0.32 3.44 7.06
    rSDt 4.325 −0.864 −1.680 −0.039 −5.398 −3.068 15.892 0.471 0.631 0.511 2.127 0.15
    3.78 −2.97 −14.57 −20.71 −3.26 −1.71 5.38 1.30 1.69 4.93 10.23
    rSWt 1.352 −0.082 −1.669 −0.016 −6.711 −2.105 2.343 0.306 0.309 0.444 1.340 0.09
    2.62 −0.42 −5.46 −6.66 −7.50 −1.85 0.59 1.56 0.83 1.69 1.13
    rTKt −1.737 0.801 −0.058 −0.015 −9.245 −2.841 82.589 1.056 −0.614 0.558 2.089 0.02
    −0.79 2.06 −5.32 −5.00 −4.54 −3.53 1.09 1.65 −2.34 4.12 6.38
    rIDt −1.443 0.796 −1.828 −0.016 −6.279 −7.660 27.982 0.473 −0.230 0.542 2.102 0.13
    −3.54 8.52 −9.92 −5.83 −2.96 −4.76 1.72 2.52 −2.97 2.63 11.42
    rKOt −3.413 1.109 −2.496 −0.028 −3.979 −3.150 26.057 0.517 −0.285 0.523 2.104 0.07
    −8.30 9.38 −9.61 −10.50 −3.54 −14.46 1.75 3.55 −1.64 2.98 5.02
    rMXt −1.542 0.849 −0.811 −0.026 −9.195 −5.491 0.022 0.091 −0.022 0.918 0.651 0.08
    −1.76 2.62 −2.32 −4.68 −5.71 −5.90 0.56 2.32 −0.13 31.88 0.83
    Notes: The lower case rt is the real stock return, rit= Ritπit for country i. πet is expected inflation, Δemvi,t is the change in equity market volatility in relation to interest rate. The numbers in the first row are the estimated coefficients, the second row contains the z-statistics. The critical values of z-distribution at the 1%, 5% and 10% levels of significance are 2.58, 1.96, and 1.65, respectively.¯R2 is the adjusted R-squared.

     | Show Table
    DownLoad: CSV

    Several findings emerge from our tests. First, Fisher's hypothesis that the real stock market returns are independent of expected inflation is mainly rejected. The evidence confirms that a negative relation between real stock market returns and expected inflation is valid and significant. The results are in line with the evidence reported by Lin (2009) and Chiang (2023). This, however, is not the case in the Japanese market where the estimated result is insignificant as low inflation and low stock return coexist (see Tables 1 and 2). On the other hand, both Brazil and Russia exhibit positive signs as these two countries experience high inflation and stock returns.

    Second, the coefficients of Δemvπ,t are negative and highly significant, indicating U.S. equity market volatility has a noticeable effect on the domestic market as well as a spillover effect to global markets. This effect appears to have a more profound effect than the countries' own conditional variance as denoted by lnσ2t. Third, for all the other estimates, the results are comparable. Therefore, the propositions are robust across different measures of stock market returns, whether it is real or nominal.

    The above analysis is based on Δemvπ,t in the estimated model, highlighting the influence of inflation on investors' sentiment that can lead to stock volatility and in turn affect stock returns. Likewise, we observe that the Fed's active attitude toward rate hikes also generates fears that cause stock market volatility. This section tests the uncertainty hypothesis via the effect of the Fed's interest rate changes on EMV (denoted by emvi,t), which could also cause the stock market to plunge (Baker et al., 2022). Table 11 reports the estimates by replacing Δemvπ,t with Δemvi,t. The statistical results are remarkable, and the qualitative directions are comparable to those using Δemvπ,t.

    Focusing on two key variables, we turn to πet and Δemvi,t in this table. First, the evidence suggests that with the exceptions of Brazil and Russia, the relation between real stock market returns and expected inflation continues to be negative and significant. This finding is consistent with the study observed by Lin (2009). Second, the coefficients of Δemvi,t are negative, and all the parameters are decidedly significant. The test results are consistent with the U.S. monetary policy stance throughout 2022–2023 when the Fed aggressively raised interest rates to stomp out inflation, causing market volatility and stock market returns to plunge. This negative effect then spread to global stock markets. This process together with the evidence that emvi,t is positively correlated with πUSt as shown in Table 4 supports the uncertainty hypothesis that runs through the channel from πUStΔemvi,tRit. It is clear that Table 11 provides solid evidence to back up the uncertainty hypothesis that postulates real stock market returns negatively respond to expected inflation.

    It is generally recognized that regression models are sensitive to sample size. When the paper was revised, more observations were available that allowed this study to extend the sample period to May 2023, which includes the later period of COVID-19 and covers more observations of the Fed's ongoing rate hikes. Due to additional observations, it is anticipated that the patten of inflation expectations might alter. For this reason, Table 12 reports the estimated coefficients for α and γ based on an adaptive expectation model. Those values were identified by the Box-Jenkins procedure. The figures reported in Table 12 are comparable, except for some minor variations for the U.S., which finds a seasonal factor. In this later case, the α values become insignificant for France and Germany. As shown in Table 6, the values of root mean square errors (RMSE) for Russia, Brazil and Turkey are relatively higher as compared with other countries, suggesting the inflation expectations are more diverse and harder to predict for these three markets.

    Table 12.  Estimates of expected inflation based on an adaptive expectation with sample period extended to May 2023.
    Expected inflation α γ RMSE
    πet(US) 0.491 0.308 0.341
    πet(CA) 0.177 0.279 0.386
    πet(UK) 0.116 0.634 0.428
    πet(FR) - 0.572 0.299
    πet(GM) - 0.442 0.382
    πet(IT) 0.267 0.329 0.281
    πet(JP) 0.196 0.418 0.358
    πet(BR) 0.904 - 3.783
    πet(RU) 0.302 - 4.800
    πet(IN) 0.387 0.398 0.8070
    πet(CN) 0.240 −0.328 0.677
    πet(SA) 0.384 0.284 0.459
    πet(ES) 0.219 0.650 0.373
    πet(NE) 0.148 0.456 0.601
    πet(SD) 0.130 0.445 0.409
    πet(SW) 0.108 0.445 0.545
    πet(TK) 0.689 0.587 1.848
    πet(ID) 0.611 - 0.971
    πet(KO) 0.368 0.411 0.455
    πet(MX) 0.808 0.398 0.491
    Notes: The expected inflation is assumed to follow an adaptive process, which is equivalent to an exponential smoothing process plus a seasonal factor. In expression, πet = πt1 + α(πt1-πet1 ) + γ. πt12,πt and πet are actual and expected inflation. RMSE is the root meaning square error.

     | Show Table
    DownLoad: CSV

    Having generated the expected inflation, the models presented in Table 11 are re-estimated and the results using a sample period of December 1989 through May 2023 are reported in Table 13. The estimated statistics presented in Table 13 are qualitatively comparable. For instance, the coefficients for both πet and Δemvi,t are negative, and only Brazil remains positive for the coefficient of expected inflation. However, the significance level for the coefficients of Δemvi,t are higher as well as the adjusted R squares. It appears that 13 out of 20 cases for the z-statistics of Δemvi,t as well as ¯R2 improved. It can be concluded that the model is robust even when the additional observations for the extended Fed's rate hikes were included.

    Table 13.  Robust estimations of real stock returns in relation to conditional variance, expected inflation, EMV calibrated to interest rate changes controlling for global financial crisis and COVID-19 using the GED-APARCH(1, 1) model : Sample 1990.01–2023.05.
    Real Returns C lnσ2t πet Δemvi,t IGFC,t ICOVID,t ω |εt1| εt1 σδt1 δ ¯R2
    rUSt 3.356 −8.631 −1.747 −0.025 −12.849 −3.620 11.082 0.120 −0.054 0.524 2.140 0.18
    2.80 −2.15 −5.45 −12.11 −49.51 −2.63 2.96 3.75 −0.60 4.47 18.95
    rCAt 1.094 −2.511 −2.625 −0.019 −8.104 −0.170 −0.485 0.597 −0.425 0.682 6.224 0.16
    8.97 −6.25 −3.94 −10.24 −8.02 −0.74 −0.04 2.99 −7.22 10.74 5.20
    rUKt 3.305 −8.988 −4.389 −0.020 −8.437 −0.984 11.863 0.139 0.011 0.563 2.073 0.11
    2.49 −1.87 −7.35 −13.25 −10.87 −3.09 1.27 3.62 4.24 5.68 5.27
    rFRt 0.239 0.175 −0.927 −0.040 −10.274 −5.984 36.613 1.004 0.697 0.853 1.763 0.14
    1.05 0.11 −2.92 −41.04 −15.45 −4.98 0.15 0.51 3.58 8.01 0.97
    rGMt 0.417 0.076 −4.814 −0.034 −9.807 −8.780 21.651 0.290 0.498 0.458 2.118 0.13
    5.32 0.35 −3.89 −11.47 −12.34 −5.92 0.19 0.74 0.69 1.19 0.85
    rITt 0.110 0.440 −3.686 −0.035 −8.665 −3.123 23.045 0.290 0.350 0.552 2.056 0.15
    0.20 0.19 −7.06 −13.27 −6.02 −7.07 0.14 0.62 0.67 1.08 0.64
    rJPt 0.610 0.297 −0.353 −0.002 0.932 0.773 0.138 0.721 −0.864 0.659 2.186 0.40
    6.60 3.62 −2.17 −12.56 21.04 14.50 5.06 4.19 −4.81 34.16 25.72
    rBRt 0.885 −2.471 0.032 −0.038 −6.297 −4.793 250.811 0.000 0.158 0.962 4.999 0.01
    0.90 −0.69 0.30 −4.69 −3.31 −2.45 2.35 −0.75 0.69 1680.64 192.11
    rRUt 0.392 0.322 −0.861 −0.026 −13.915 −14.136 0.373 0.180 −0.224 0.776 0.783 0.08
    0.55 0.15 −0.85 −3.57 −5.28 −9.43 0.94 2.61 −1.21 11.25 1.40
    rINt 0.263 −0.021 −0.379 −0.024 −5.147 −4.394 11.436 0.544 −0.062 0.766 2.217 0.04
    4.92 −0.13 −2.61 −10.72 −1.93 −6.72 0.22 1.13 −0.26 5.30 1.23
    rCNt −1.066 5.481 −1.455 −0.038 −11.621 0.444 70.898 0.598 −0.005 0.521 2.073 0.03
    −1.49 1.46 −4.13 −13.32 −5.47 0.37 1.08 1.84 −1.23 2.96 12.77
    rSAt 0.244 1.310 −2.319 −0.037 −6.808 −3.026 14.995 0.358 0.572 0.608 2.399 0.17
    1.74 2.60 −14.77 −33.14 −5.10 −9.55 0.57 0.72 0.66 5.60 2.40
    rESt −0.125 1.292 −1.042 −0.035 −9.906 1.337 13.222 0.181 0.190 0.578 2.073 0.10
    −1.63 3.53 −2.95 −14.81 −8.39 0.76 1.08 0.77 0.34 1.56 3.82
    rNEt −0.470 3.281 −4.548 −0.033 −5.614 −4.058 8.581 0.322 0.254 0.524 2.170 0.18
    −1.63 3.80 −5.67 −12.98 −3.09 −2.83 2.73 1.75 1.50 3.70 5.73
    rSDt −0.881 3.621 −2.076 −0.040 −6.766 −6.559 21.637 0.152 0.063 0.598 2.003 0.12
    −0.63 0.62 −7.60 −22.57 −4.75 −2.89 0.35 0.51 0.61 1.25 1.58
    rSWt −0.296 1.668 −0.567 −0.029 −7.681 −4.216 12.475 0.454 0.329 0.512 2.093 0.13
    −1.19 1.92 −1.79 −16.96 −6.19 −2.73 1.50 1.05 1.22 2.16 4.15
    rTKt −2.456 12.676 −0.087 −0.035 −6.450 −3.788 67.774 0.460 0.199 0.535 2.135 0.03
    −1.95 1.95 −1.75 −8.62 −2.62 −1.72 0.72 1.81 3.19 5.45 3.69
    rIDt 0.788 −1.285 −1.059 −0.013 −10.687 −6.496 3.289 0.188 −0.141 0.838 2.620 0.12
    3.44 −1.73 −3.88 −3.41 −4.85 −3.94 0.64 1.53 −0.78 11.64 2.38
    rKOt 0.570 −0.561 −4.031 −0.052 −11.085 −5.561 7.215 0.363 −0.046 0.812 2.201 0.08
    2.15 −0.49 −9.92 −16.07 −13.56 −6.81 0.29 1.29 −0.25 6.93 1.39
    rMXt 0.854 −1.174 −0.827 −0.012 −9.566 −5.581 0.140 0.065 −0.213 0.944 2.566 0.05
    3.26 −1.52 −2.73 −3.18 −12.74 −5.52 0.39 1.21 −1.11 37.98 1.44
    Notes: The lower case rt is the real stock return, rit= Ritπit for country i. πet is expected inflation, Δemvi,t is the change in equity market volatility in relation to interest rate. The numbers in the first row are the estimated coefficients, the second row contains the z-statistics. The critical values of z-distribution at the 1%, 5% and 10% levels of significance are 2.58, 1.96, and 1.65, respectively.¯R2 is the adjusted R-squared.

     | Show Table
    DownLoad: CSV

    Since the evidence in Table 4 shows that equity market volatilities, whether calibrated to inflation or change in interest rates, are positively correlated with the implied volatility and inflation, they have an effect on the real stock market returns. In addition, Yang et al. (2006) note that U.S. inflation has a direct effect on inflation in global markets. It is, therefore, of interest to conduct a direct test that links real stock returns to domestic and U.S. expected inflation. The statistical results are provided in Table 14.

    Table 14.  Estimates of real stock returns in relation to conditional variance, domestic expected inflation and U.S. expected inflation controlling for global financial crisis and COVID-19 using the GED-APARCH(1, 1) model.
    Real Returns C lnσ2t πe,it πe,USt IGFC,t ICOVID,t ω |εt1| εt1 σδt1 δ ¯R2
    rUSt −6.808 3.000 −0.655 −4.765 −5.852 −2.920 9.088 0.100 −0.154 0.562 2.158 0.11
    −7.49 15.58 −4.44 −7.97 −26.58 −8.67 9.22 5.28 −5.32 6.59 17.88
    rCAt −9.338 3.625 −1.627 −5.713 −9.678 −2.255 12.414 0.114 −0.149 0.530 2.117 0.05
    −2.22 2.88 −8.03 −7.30 −8.06 −1.68 1.60 2.97 −5.45 4.22 7.17
    rUKt −8.516 3.424 −1.947 −3.416 −5.384 −1.726 10.404 0.125 −0.263 0.532 2.139 0.05
    −2.75 3.67 −9.57 −4.18 −5.01 −1.42 5.34 3.95 −2.48 5.79 18.93
    rFRt −2.982 1.502 −0.825 −4.028 −8.084 −2.761 16.618 −0.027 −0.427 0.552 2.070 0.02
    −0.25 0.45 −2.67 −3.87 −7.25 −2.50 0.36 −0.63 −0.65 0.47 2.81
    rGMt 0.517 0.698 −0.960 −6.438 −8.863 −3.748 1.689 0.203 0.045 0.646 1.332 0.04
    0.35 1.46 −1.84 −4.27 −6.10 −2.98 0.51 1.54 0.15 3.20 1.12
    rITt −8.525 2.827 −1.815 −7.830 −8.723 0.532 23.698 0.040 −0.282 0.550 2.064 0.04
    −0.47 0.61 −2.84 −6.27 −5.99 0.74 0.21 0.64 −1.08 1.24 0.89
    rJPt 66.391 −19.276 −1.281 −8.291 −6.004 −0.648 17.904 0.016 0.749 0.528 2.198 0.01
    5.49 −4.91 −2.30 −4.62 −3.16 −0.40 4.48 0.65 0.64 5.93 18.67
    rBRt 0.195 0.447 0.173 −6.884 −8.043 −5.124 0.823 −0.003 −0.209 0.994 2.711 0.01
    0.05 0.44 2.24 −1.97 −4.02 −2.40 0.65 −1.03 −0.26 3525.1 3.09
    rRUt 9.896 −1.519 0.736 −14.577 −4.114 −1.839 49.752 0.239 0.462 0.462 2.159 0.05
    2.91 −2.03 3.58 −6.73 −1.81 −0.94 1.05 1.38 1.74 2.76 4.41
    rINt −0.266 0.796 −1.011 −7.053 −15.120 −9.001 0.807 0.101 −0.361 0.885 2.580 0.03
    −0.19 1.86 −2.98 −2.20 −8.99 −16.46 0.58 3.27 −3.37 38.42 3.50
    rCNt −2.393 0.828 −0.849 −4.446 −9.897 −3.335 31.491 1.796 −0.303 0.532 2.094 0.02
    −1.28 2.20 −5.07 −11.43 −4.23 −4.69 2.63 1.59 −5.05 3.26 7.17
    rSAt 0.338 0.628 −2.018 −2.544 −6.490 −2.965 15.131 0.185 0.272 0.549 2.100 0.07
    0.26 1.77 −8.55 −6.96 −6.23 −1.37 0.26 0.64 0.38 1.47 1.15
    rESt −10.404 3.404 −2.054 −4.049 −10.132 −8.968 18.150 0.118 −0.140 0.563 2.116 0.05
    −1.22 1.51 −8.93 −2.80 −9.15 −3.54 3.44 1.58 −0.91 3.19 13.30
    rNEt 0.070 0.577 −0.820 −4.002 −7.387 −1.851 16.460 0.193 0.199 0.582 2.066 0.04
    1.15 10.40 −5.10 −5.81 −6.00 −3.62 0.85 1.18 1.27 3.13 3.98
    rSDt 4.325 −0.864 −1.680 −0.039 −5.398 −3.068 15.892 0.471 0.631 0.511 2.127 0.15
    3.78 −2.97 −14.57 −20.71 −3.26 −1.71 5.38 1.30 1.69 4.93 10.23
    rSWt 1.352 −0.082 −1.669 −0.016 −6.711 −2.105 2.343 0.306 0.309 0.444 1.340 0.09
    2.62 −0.42 −5.46 −6.66 −7.50 −1.85 0.59 1.56 0.83 1.69 1.13
    rTKt −3.315 1.777 −0.195 −9.619 −11.109 −1.677 1.175 0.057 −0.403 0.923 1.923 0.02
    −1.34 2.90 −0.33 −2.20 −4.89 −0.61 0.25 1.24 −1.01 23.63 1.12
    rIDt −0.565 0.902 −1.634 −5.097 −7.126 −6.025 33.661 0.468 −0.629 0.587 2.034 0.09
    −0.57 4.43 −8.34 −5.33 −3.29 −2.79 0.59 1.20 −1.47 3.12 2.53
    rKOt −1.476 1.030 −1.821 −9.446 −5.732 −1.744 21.217 0.251 −1.000 0.636 2.206 0.07
    −2.09 5.38 −10.60 −20.96 −4.87 −3.46 0.95 2.89 −4.70 9.94 4.02
    rMXt −11.123 3.403 −1.366 −4.344 −6.258 0.273 25.630 0.120 −0.968 0.533 2.084 0.06
    −3.60 4.54 −5.72 −2.69 −4.74 0.29 0.87 0.15 −0.15 4.34 3.79
    Note: This table is based on the period January 1990 – June 2022. The lower case rt is the real stock return. πe,it and πe,,USt are domestic. expected inflation and U.S. expected inflation, respectively. The UK inflation was used in estimating the US market because the UK market shares a similar feature to that of the US market. The numbers in the first row are the estimated coefficients, the second row contains the z-statistics. The critical values of t-distribution at the 1%, 5% and 10% levels of significance are 2.58, 1.96, and 1.65, respectively.¯R2 is the adjusted R-squared.

     | Show Table
    DownLoad: CSV

    The evidence indicates that the model is robust across different sources of expected inflation. Specifically, domestic expected inflation produces very comparable estimated results as the previous tables indicate, the exceptions are Brazil and Russia, which have positive coefficients. In particular, the coefficients of πet for the other 18 countries show significantly negative signs and are consistent with previous findings (Fama, 1981; James et al., 1985; Kolluri and Wahab, 2008). Even more remarkable is the result that all coefficients of πe,USt are negative and significant at the conventional level, including the coefficients for Brazil and Russia. Moreover, the magnitude of coefficients for the absolute value is much larger in each country, indicating the profound impact of U.S. inflation on global stock returns.

    Fama's (1981) proxy hypothesis provides the basis for testing the relation between stock market returns and inflation. His interpretation (1981) is based on the premise there is a negative relation between inflation and real activity and a positive correlation between real activity and stock market returns. Geske and Roll (1983) observe that there is a negative effect from a rise in expected inflation on stock market returns, which can be achieved from reversed money growth/tax revenues, which shifts inflation downward. While both hypotheses (Fama, 1981 and Geske and Roll, 1983) are valid in a regime of low or moderate inflationary environment when inflation expectations do not impinge on economic agents' sentiment. We provide an alternative explanation that focuses on the premise that inflation could trigger investors' fears, causing increased equity market volatility and stock returns to plummet.

    Testing of major stock markets using monthly data shows that the stock returns of 18 out of 20 examined countries (the exceptions being Brazil and Russia) are negatively correlated to (expected) inflation. The evidence of a negative correlation is consistent with the hypotheses of Modigliani and Cohn (1979), Fama (1981), Geske and Roll (1983) and McCarthy et al. (1990). However, we found that the evidence consistently supports a negative relation between (real) stock market returns and (expected) inflation, which passes through two different channels: (a) The U.S. EMV has a negative impact on U.S. stock returns as well as a spillover effect to global markets and (b) EMV is positively related to U.S. inflation or interest rate changes. Thus, these two connections provide evidence to substantiate the uncertainty hypothesis that holds a negative relation between stock market returns and inflation. The finding is consistent with the result reported by Chiang (2023).

    The results are robust utilizing different measures of different sample periods, inflation expectations, domestic and U.S. expected inflations, real stock returns vs. the nominal stock returns and equity market volatility induced by inflation, which causes a change in investors' sentiments or is driven by monetary policy uncertainty as reflected by the Fed's behavior.

    Finally, while conducting the empirical estimations, the extreme observations during the periods of the 2008-2009 global financial crisis and the 2020–2021 pandemic are found to have significant downturn effects. Using dummy variables to control these two extreme time periods helps to mitigate the estimated biases. Further, using the GED-APARCH(1, 1) procedure in this study is appealing, since the model is able to address the volatility clustering, fat-tail, asymmetric shock and long memory property for the stock return series.

    This study has significant implications as it draws particular attention to the impact of inflation induced equity market volatility, which can change investors' sentiment and affect stock returns. In addition, this study indicates that the Fed's monetary policy uncertainty on interest rates can cause equity market volatility that causes the stock market to plunge. All these uncertainty factors should be priced into the stock investments.

    The authors declare they have not used Artificial Intelligence (AI) tools in the creation of this article.

    All authors declare no conflicts of interest in this paper.



    [1] A. P. Plageras, K. E. Psannis, C. Stergiou, H. Wang, B. B. Gupta, Efficient IoT-based sensor BIG Data collection–processing and analysis in smart buildings. Future Gener Comput Syst, 82 (2018), 349–357. https://doi.org/10.1016/j.future.2017.09.082 doi: 10.1016/j.future.2017.09.082
    [2] Z. Ahmed, M. Ayaz, M. A. Hijji, M. Z. Abbas, A. Rahim, AUV-Based efficient data collection scheme for underwater linear sensor networks. Int J Semant Web Inf Syst, 18 (2022), 1–19. https://doi.org/10.4018/IJSWIS.299858 doi: 10.4018/IJSWIS.299858
    [3] D. Tian, Y. Han, B. Wang, T. Guan, H. Gu, W. Wei, Review of object instance segmentation based on deep learning, J. Electron. Imag., 31 (2021), 041205. https://doi.org/10.1117/1.JEI.31.4.041205 doi: 10.1117/1.JEI.31.4.041205
    [4] K. S. Arikumar, A. D. Kumar, T. R. Gadekallu, S. B. Prathiba, K. Tamilarasi, Real-Time 3D Object detection and classification in autonomous driving environment using 3D LiDAR and camera sensors, Electronics, 11 (2022), 4203. https://doi.org/10.3390/electronics11244203 doi: 10.3390/electronics11244203
    [5] H. Song, A more efficient approach for remote sensing image classification, Comput. Mater. Contin., 74 (2023), 5741–5756. https://doi.org/10.32604/cmc.2023.034921 doi: 10.32604/cmc.2023.034921
    [6] H. Song, FST-EfficientNetV2: exceptional image classification for remote sensing, Comput. Sci. Eng., 46 (2023), 3959–3978. https://doi.org/10.32604/csse.2023.038429 doi: 10.32604/csse.2023.038429
    [7] H. Touvron, A. Vedaldi, M. Douze, H. Jégou, Fixing the train-test resolution discrepancy, arXiv: 1906.06423, [preprint], (2019)[cited 2023 September 05]. Available from: http://arXiv.org/abs/1906.06423
    [8] H. Song, A Leading but Simple Classification Method for Remote Sensing Images, AETiC, 7 (2023), 1–20. https://doi.org/10.33166/AETiC.2023.03.001 doi: 10.33166/AETiC.2023.03.001
    [9] M. Tan, Q. V. Le, EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks, arXiv: 1905.11946, [preprint], (2019)[cited 2023 September 05]. Available from: http://arXiv.org/abs/1905.11946
    [10] G. Cheng, C. Yang, X. Yao, L. Guo, J. Han, When deep learning meets metric learning: remote sensing image scene classification via learning discriminative CNNs, IEEE Trans Geosci Remote Sens, 56 (2018), 2811–2821. https://doi.org/10.1109/TGRS.2017.2783902 doi: 10.1109/TGRS.2017.2783902
    [11] Y. Guo, J. Ji, X. Lu, H. Huo, T. Fang, D. Li, Global-Local attention network for aerial scene classification, IEEE Access, 7 (2019), 67200–67212. https://doi.org/10.1109/ACCESS.2019.2918732 doi: 10.1109/ACCESS.2019.2918732
    [12] R. Minetto, M. P. Segundo, S. Sarkar, Hydra: An ensemble of convolutional neural networks for geospatial land classification, IEEE Trans Geosci Remote Sens, 57 (2019), 6530–6541. https://doi.org/10.1109/TGRS.2019.2906883 doi: 10.1109/TGRS.2019.2906883
    [13] W. Tong, W. Chen, W. Han, X. Li, L. Wang, Channel-Attention-Based DenseNet network for remote sensing image scene classification, IEEE J Sel Top Appl Earth Obs Remote Sens, 13 (2020), 4121–4132. https://doi.org/10.1109/JSTARS.2020.3009352 doi: 10.1109/JSTARS.2020.3009352
    [14] J. Li, D. Lin, Y. Wang, G. Xu, C. Ding, Deep discriminative representation learning with attention map for scene classification, arXiv: 1902.07967, [preprint], (2019)[cited 2023 September 05]. Available from: http://arXiv.org/abs/1902.07967
    [15] H. Alhichri, A. S. Alswayed, Y. Bazi, N. Ammour, N. A. Alajlan, Classification of remote sensing images using efficientnet-b3 cnn model with attention, IEEE Access, 9 (2021), 14078–14094. https://doi.org/10.1109/ACCESS.2021.3051085 doi: 10.1109/ACCESS.2021.3051085
    [16] B. Li, Y. Guo, J. Yang, L. Wang, Y. Wang, W. An, Gated recurrent multiattention network for VHR remote sensing image classification, IEEE Trans Geosci Remote Sens, 60 (2022), 1–13. https://doi.org/10.1109/TGRS.2021.3093914 doi: 10.1109/TGRS.2021.3093914
    [17] W. Chen, S. Ouyang, W. Tong, X. Li, X. Zheng, L. Wang, GCSANet: A global context spatial attention deep learning network for remote sensing scene classification, IEEE J Sel Top Appl Earth Obs Remote Sens, 15 (2022), 1150–1162. https://doi.org/10.1109/JSTARS.2022.3141826 doi: 10.1109/JSTARS.2022.3141826
    [18] Q. Zhao, Y. Ma, S. Lyu, L. Chen, Embedded Self-Distillation in compact multibranch ensemble network for remote sensing scene classification, IEEE Trans Geosci Remote Sens, 60 (2022), 1–15. https://doi.org/10.1109/TGRS.2021.3126770 doi: 10.1109/TGRS.2021.3126770
    [19] H. Song, A consistent mistake in remote sensing images' classification literature, Intell. Autom. Soft Comput., 37 (2023), 1381–1398. https://doi.org/10.32604/iasc.2023.039315 doi: 10.32604/iasc.2023.039315
    [20] S. Chaib, H. Liu, Y. Gu, H. Yao, Deep feature fusion for vhr remote sensing scene classification, IEEE Trans Geosci Remote Sens, 55 (2017), 4775–4784. https://doi.org/10.1109/TGRS.2017.2700322 doi: 10.1109/TGRS.2017.2700322
    [21] Y. Liu, C. Y. Suen, Y. Liu, L. Ding, Scene classification using hierarchical wasserstein CNN, IEEE Trans Geosci Remote Sens, 57 (2019), 2494–2509. https://doi.org/10.1109/TGRS.2018.2873966 doi: 10.1109/TGRS.2018.2873966
    [22] Y. Liu, Y. Liu, L. Ding, Scene classification by coupling convolutional neural networks with wasserstein distance, IEEE Geosci. Remote Sensing Lett., 16 (2019), 722–726. https://doi.org/10.1109/LGRS.2018.2883310 doi: 10.1109/LGRS.2018.2883310
    [23] Y. Bazi, M. M. Al Rahhal, H. Alhichri, N. Alajlan, Simple yet effective fine-tuning of deep CNNs using an auxiliary classification loss for remote sensing scene classification, Remote Sensing, 11 (2019), 2908. https://doi.org/10.3390/rs11242908 doi: 10.3390/rs11242908
    [24] W. Zhang, P. Tang, L. Zhao, Remote sensing image scene classification using CNN-CapsNet, Remote Sensing, 11 (2019), 494. https://doi.org/10.3390/rs11050494 doi: 10.3390/rs11050494
    [25] J. Xie, N. He, L. Fang, A. Plaza, Scale-Free convolutional neural network for remote sensing scene classification, IEEE Trans Geosci Remote Sens, 57 (2019), 6916–6928. https://doi.org/10.1109/TGRS.2019.2909695 doi: 10.1109/TGRS.2019.2909695
    [26] H. Sun, S. Li, X. Zheng, X. Lu, Remote sensing scene classification by gated bidirectional network, IEEE Trans Geosci Remote Sens, 58 (2020), 82–96. https://doi.org/10.1109/TGRS.2019.2931801 doi: 10.1109/TGRS.2019.2931801
    [27] D. Guo, Y. Xia, X. Luo, Scene classification of remote sensing images based on saliency dual attention residual network, IEEE Access, 8 (2020), 6344–6357. https://doi.org/10.1109/ACCESS.2019.2963769 doi: 10.1109/ACCESS.2019.2963769
    [28] X. Tang, Q. Ma, X. Zhang, F. Liu, J. Ma, L. Jiao, Attention consistent network for remote sensing scene classification, IEEE J Sel Top Appl Earth Obs Remote Sens, 14 (2021), 2030–2045. https://doi.org/10.1109/JSTARS.2021.3051569 doi: 10.1109/JSTARS.2021.3051569
    [29] Y. Bazi, L. Bashmal, M. M. A. Rahhal, R. A. Dayil, N. A. Ajlan, Vision transformers for remote sensing image classification, Remote Sensing, 13 (2021), 516. https://doi.org/10.3390/rs13030516 doi: 10.3390/rs13030516
    [30] J. Zhang, H. Zhao, J. Li, TRS: transformers for remote sensing scene classification, Remote Sensing, 13 (2021), 4143. https://doi.org/10.3390/rs13204143 doi: 10.3390/rs13204143
    [31] D. Wang, J. Zhang, B. Du, G. S. Xia, D. Tao, An empirical study of remote sensing pretraining, arXiv: 2204.02825, [preprint], (2022)[cited 2023 September 06]. Available from: http://arXiv.org/abs/2204.02825
    [32] C. Shi, X. Zhang, J. Sun, L. Wang, Remote sensing scene image classification based on Self-Compensating convolution neural network, Remote Sensing, 14 (2022), 545. https://doi.org/10.3390/rs14030545 doi: 10.3390/rs14030545
    [33] S. B. Chen, Q. S. Wei, W. Z. Wang, J. Tang, B. Luo, Z. Y. Wang, Remote sensing scene classification via Multi-Branch local attention network, IEEE Trans. on Image Process., 31 (2022), 99–109. https://doi.org/10.1109/TIP.2021.3127851 doi: 10.1109/TIP.2021.3127851
    [34] P. Deng, K. Xu, H. Huang, When CNNs meet vision transformer: a joint framework for remote sensing scene classification, IEEE Geosci. Remote Sensing Lett., 19 (2022), 1–5. https://doi.org/10.1109/LGRS.2021.3109061 doi: 10.1109/LGRS.2021.3109061
    [35] W. Miao, J. Geng, W. Jiang, Multigranularity decoupling network with pseudolabel selection for remote sensing image scene classification, IEEE Trans Geosci Remote Sens, 61 (2023), 1–13. https://doi.org/10.1109/TGRS.2023.3244565 doi: 10.1109/TGRS.2023.3244565
    [36] H. Song, W. Yang, GSCCTL: a general semi-supervised scene classification method for remote sensing images based on clustering and transfer learning, Int J Remote Sens., 43 (2022), 5976–6000. https://doi.org/10.1080/01431161.2021.2019851 doi: 10.1080/01431161.2021.2019851
    [37] W. Wang, Y. Chen, P. Ghamisi, Transferring CNN with adaptive learning for remote sensing scene classification, IEEE Trans Geosci Remote Sens, 60 (2022), 1–18. https://doi.org/10.1109/TGRS.2022.3190934 doi: 10.1109/TGRS.2022.3190934
    [38] K. Xu, P. Deng, H. Huang, Vision transformer: an excellent teacher for guiding small networks in remote sensing image scene classification, IEEE Trans Geosci Remote Sens, 60 (2022), 1–15. https://doi.org/10.1109/TGRS.2022.3152566 doi: 10.1109/TGRS.2022.3152566
    [39] T. He, Z. Zhang, H. Zhang, Z. Zhang, J. Xie, M. Li, Bag of tricks for image classification with convolutional neural networks, Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, (2019), 558–567.
    [40] S. Yun, D. Han, S. J. Oh, S. Chun, J. Choe, Y. Yoo, CutMix: Regularization strategy to train strong classifiers with localizable features, Proceedings of the IEEE/CVF international conference on computer vision, (2019), 6023–6032.
    [41] C. B. Zhang, P. T. Jiang, Q. Hou, Y. Wei, Q. Han, Z. Li, et al., Delving deep into label smoothing, IEEE Trans. on Image Process., 30 (2021), 5984–5996. https://doi.org/10.1109/TIP.2021.3089942 doi: 10.1109/TIP.2021.3089942
    [42] J. Hu, L. Shen, S. Albanie, G. Sun, E. Wu, Squeeze-and-Excitation Networks, Proceedings of the IEEE conference on computer vision and pattern recognition, (2018), 7132–7141.
    [43] R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, D. Batra, Grad-CAM: visual explanations from deep networks via Gradient-Based localization, Proceedings of the IEEE international conference on computer vision, (2017), 618–626.
    [44] I. Loshchilov, F. Hutter, Decoupled weight decay regularization, arXiv: 1711.05101v3, [preprint], (2017)[cited 2023 September 05]. Available from: http://arXiv.org/abs/1711.05101v3
    [45] C. Szegedy, V. Vanhoucke, S. Ioffe, J. Shlens, Z. Wojna, Rethinking the inception architecture for computer vision, Proceedings of the IEEE conference on computer vision and pattern recognition, 2016, 2818–2826.
    [46] H. Zhang, M. Cisse, Y. N. Dauphin, D. Lopez-Paz, Mixup: beyond empirical risk minimization, arXiv: 1710.09412, [preprint], (2017)[cited 2023 September 05]. Available from: http://arXiv.org/abs/1710.09412
    [47] Y. H. Liu, E. Sangineto, W. Bi, N. Sebe, B. Lepri, M. Nadai, Efficient training of visual transformers with small datasets, NIPS, 34 (2021), 23818–23830.
    [48] A. Hassani, S. Walton, N. Shah, A. Abuduweili, J. Li, H. Shi, Escaping the big data paradigm with compact transformers, arXiv: 2104.05704, [preprint], (2021)[cited 2023 September 05]. Available from: http://arXiv.org/abs/2104.05704
    [49] Z. Li, F. Liu, W. Yang, S. Peng, J. Zhou, A survey of convolutional neural networks: analysis, applications, and prospects, IEEE Trans. Neural Netw. Learning Syst., 33 (2022), 6999–7019. https://doi.org/10.1109/TNNLS.2021.3084827 doi: 10.1109/TNNLS.2021.3084827
    [50] F. Liu, D. Chen, Z. Guan, X. Zhou, J. Zhu, J. Zhou, RemoteCLIP: A vision language foundation model for remote sensing, arXiv: 2306.11029, [preprint], (2023)[cited 2023 September 05]. Available from: http://arXiv.org/abs/2306.11029
    [51] L van der Maaten, G. Hinton, Visualizing data using t-SNE, J Mach Learn Res, 9 (2008), 2579–2605.
    [52] Y. Long, G. S. Xia, S. Li, W. Yang, M. Y. Yang, X. X. Zhu, et al., On creating benchmark dataset for aerial image interpretation: reviews, guidances, and Million-AID, IEEE J Sel Top Appl Earth Obs Remote Sens, 14 (2021), 4205–4230. https://doi.org/10.1109/JSTARS.2021.3070368 doi: 10.1109/JSTARS.2021.3070368
  • This article has been cited by:

    1. Malika Neifar, Amira Harzallah, The Fisher’s hedge hypothesis: what about homogeneity and stability properties?, 2024, 1229-988X, 10.1108/JDQS-07-2024-0028
    2. Jiajia He, Xiuping Zou, Tinghui Li, House price, gender spatial allocation, and the change of marriage matching, 2024, 9, 2473-6988, 8079, 10.3934/math.2024393
    3. Zhongzhe Ouyang, Ke Liu, Min Lu, Bias correction based on AR model in spurious regression, 2024, 9, 2473-6988, 8439, 10.3934/math.2024410
    4. Austin Murphy, Zeina AlSalman, Ioannis Souropanis, An Investigation into the Causes of Stock Market Return Deviations from Real Earnings Yields, 2025, 10590560, 104216, 10.1016/j.iref.2025.104216
  • Reader Comments
  • © 2023 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(1764) PDF downloads(150) Cited by(10)

Figures and Tables

Figures(11)  /  Tables(7)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog