Skip to main content

Table 8 Estimation results using news-based economic sentiment index

From: Can news-based economic sentiment predict bubbles in precious metal markets?

 

Dependent variable: Bubble

Gold

Silver

Palladium

Platinum

Panel A: Probit models

\(NESI_{t - 1}\)

− 1.5088***

0.2503

0.7318

− 1.0746***

(0.0020)

(0.7400)

(0.2810)

(0.0000)

\(Inflation_{t - 1}\)

0.3800**

0.1281*

0.5802***

0.2080***

(0.0200)

(0.0940)

(0.0000)

(0.0000)

\(US{\text{DI}}_{t - 1}\)

− 1.2551***

− 0.2018**

− 0.0546***

− 0.6100***

(0.0030)

(0.0120)

(0.0080)

(0.0021)

\({\text{EFR}}_{t - 1}\)

− 0.1560**

0.2178

− 0.4890***

− 0.2283***

(0.0240)

(0.2110)

(0.0000)

(0.0000)

\(T - Spread_{t - 1}\)

− 0.4356*

− 1.2005**

− 0.3744**

− 0.4551***

(0.0780)

(0.0120)

(0.0311)

(0.0200)

\({\text{GEA}}_{t - 1}\)

0.0102***

0.0060**

0.0045**

0.0062***

(0.0000)

(0.0200)

(0.0138)

(0.0000)

\(Constant\)

0.3189***

0.3698***

0.5793***

0.2435***

(0.0000)

(0.0029)

(0.0000)

(0.0000)

Panel B: Conditional marginal effects

\(NESI_{t - 1}\)

− 0.2675***

0.0170

0.0973

− 0.2425***

(0.0060)

(0.7430)

(0.2440)

(0.0010)

\(Inflation_{t - 1}\)

0.2447**

0.0871*

0.7419***

0.0498***

(0.0360)

(0.0945)

(0.0000)

(0.0000)

\(US{\text{DI}}_{t - 1}\)

− 0.2225***

− 0.0137***

− 0.0073***

− 0.0226***

(0.0010)

(0.0020)

(0.0010)

(0.0020)

\({\text{EFR}}_{t - 1}\)

− 0.0277**

0.0148

− 0.0650***

− 0.0515***

(0.0120)

(0.2070)

(0.0000)

(0.0000)

\(T - Spread_{t - 1}\)

− 0.0772**

− 0.0816**

− 0.0498**

− 0.1027**

(0.0750)

(0.0160)

(0.0292)

(0.0220)

\({\text{GEA}}_{t - 1}\)

0.0018***

0.0004**

0.0006**

0.0014***

(0.0000)

(0.0330)

(0.0135)

(0.0000)

\(Observations\)

420

420

420

420

McFadden's pseud-R2

0.8334

0.5714

0.4757

0.6726

Log-likelihood

− 117.6715

− 48.2477

− 92.6440

− 162.2984

Hosmer–Lemeshow test

7.09

8.12

6.67

11.97

(0.3690)

(0.1887)

(0.1598)

(0.2176)

\({\text{Correct bubble}}\)

84.13%

75.00%

74.42%

96.30%

\({\text{Correct no}} - {\text{bubble}}\)

96.99%

96.21%

91.67%

86.88%

\({\text{Correct }}\left( {{\text{classified}}} \right){\text{ overall}}\)

92.36%

95.59%

91.91%

87.50%

  1. The dependent variable is a binary that equals 1 (bubble dates) and 0 (none-bubble dates) identified by the GSADF procedure. Panels A and B report the results of the probit regressions and conditional marginal effects of a unit change in the mean value of the explanatory variables on the probability of a bubble. The Hosmer–Lemeshow test is a statistical test for goodness of fit for probit regressions, which follows an \(\chi^{2}\) distribution. A large \(\chi^{2}\) value (with small p value \(< 0.05\)) indicates poor fit regression model. The last three (bottom) rows show the percentage of bubbles that are correctly identified at predicted probability \(> 0.5 \left( {50\% } \right)\). Robust standard errors are given in parentheses. p values are given in brackets.*, **, and *** indicate significance at 10%, 5%, and 1% levels, respectively.