Do cryptocurrencies exhibit similar hedging and safe haven
properties under different uncertainty indices? A comparative
analysis with gold and the U.S. Dollar index
Pranay Kumar Ankam
*
, Jamini Kanta Pattanayak
Department of Management Studies and Industrial Engineering, Indian Institute of Technology (Indian School of Mines) Dhanbad, Jharkhand, India
ARTICLE INFO
Keywords:
Cryptocurrency ETFs
Hedging
Safe-haven
GARCH
Wavelet quantile correlation
ABSTRACT
This study investigates hedging and safe-haven properties of cryptocurrency indices (S&P Bitcoin,
S&P Broad digital market) and traditional safe-havens (Gold, U.S. Dollar index) under various
uncertainty sources using GARCH and Wavelet quantile correlation methods. Uncertainty sources
include CBOE Volatility Index (VIX), CBOE Emerging markets ETF volatility Index (VXEEM),
CBOE EFA volatility Index (VXEFA), CBOE Crude oil Volatility Index (OVX), Economic policy
uncertainty (EPU), and Geopolitical risk (GPR). Our findings indicate that hedging and safe-haven
properties of cryptocurrency indices and traditional safe-havens are heterogenous under different
sources of uncertainty. The results hold true even after decomposing the returns into various
quantiles and frequencies.
1. Introduction
The global economies underwent multiple shocks, including the U.S. trade tariffs, geopolitical tensions, Fed policies, and the Covid-
19 pandemic and its aftermath. As a result we observe frequent stock market volatility jumps, fluctuations in crude oil prices, and
economic instability across the countries. Under these circumstances, investors are keen to know which asset may effectively protect
from these numerous shocks and reduce portfolio variance. Built on portfolio optimisation theory and safe-haven hypothesis (Baur and
Lucey, 2010), this study examines whether investors should prefer cryptocurrencies (Bouri et al., 2017) or traditional safe-havens such
as gold (Akhtaruzzaman et al., 2021) or currencies (Park and Fang, 2025) against shocks and the resulting uncertainties.
Although, highly volatile and speculative (Ali et al., 2025b), cryptocurrencies’ decentralization, low asset correlation, and digital
gold status indicate their hedge and safe-haven abilities (Chibane and Janson, 2024). Moreover, Bitcoin’s price surge beyond 100,000$
and cryptocurrency ETF launches has further accelerated investor interest.
Existing studies assert that cryptocurrencies provide limited hedging and safe-haven roles across stock market uncertainty (VIX)
and economic policy uncertainty (EPU) (Bouri et al., 2017; Demir et al., 2018). In contrast, Long et al. (2021) show that they cannot
serve as safe-haven when faced with uncertainties like VIX, OVX (oil market uncertainty), and EPU, possibly due to liquidity risk
(Zhang et al., 2023). On the other hand, gold consistently shows strong safe-haven behaviour across stock markets and policy related
uncertainties by maintaining stability during economic downturns and inflationary periods (Past´
en-Henríquez et al., 2025; Wu et al.,
2019). Additionally, U.S. dollar performs as a consistent hedge against stock markets (Sharma and Karmakar, 2022) nevertheless
* Corresponding author at: Department of Management Studies and Industrial Engineering, Indian Institute of Technology (Indian School of
Mines) Dhanbad, Jharkhand 826004, India.
E-mail addresses: 21dr0022@ms.iitism.ac.in (P.K. Ankam), jkpattanayak@iitism.ac.in (J.K. Pattanayak).
Contents lists available at ScienceDirect
Finance Research Letters
journal homepage: www.elsevier.com/locate/frl
https://doi.org/10.1016/j.frl.2025.108934
Received 26 August 2025; Received in revised form 6 November 2025; Accepted 11 November 2025
Finance Research Letters 86 (2025) 108934
Available online 12 November 2025
1544-6123/© 2025 Elsevier Inc. All rights are reserved, including those for text and data mining, AI training, and similar technologies.
rarely analysed across EPU and geopolitical risk (GPR). Further, Bitcoin and gold’s safe-haven properties rely either on market con-
ditions or investment horizons (Selmi et al., 2021). Critically, we observe that cryptocurrencies, gold and U.S. dollar hedging and
safe-haven abilities are analysed against VIX, OVX, GPR, and EPU indices in isolation (Das et al., 2019; Mo et al., 2025), with only few
studies examining them simultaneously (Colon et al., 2020). There are no prior studies that jointly compared cryptocurrencies, gold,
and U.S. dollar index assets response to market-implied (VIX and OVX) and event-driven (EPU and GPR) uncertainties.
This study fills this gap by comparing the hedging and safe-haven properties of S&P Bitcoin index (SPBTC), S&P Broad Digital
Market index (SPBDCI), gold and U.S. Dollar index against stock market volatility indices (VIX, VXEEM, VXEFA), crude oil volatility
(OVX), U.S. economic policy (EPU), and geopolitical risk (GPR) uncertainty indices. We focus on SPBDCI for its diversified behaviour
and idiosyncratic risk benefits. Methodologically, we utilize GARCH (1,1) model to examine assets hedging and safe-haven behaviour
under different uncertainty shocks. However, due to its limitations in capturing multi-scale and quantile-dependent relationships, we
incorporate Wavelet Quantile Correlation (WQC) method enhanced by bootstrapping procedure. Our results do not indicate direct
trading or portfolio losses, as we focus on uncertainties, not on investable securities. We rather aim to analyse how these assets respond
to different uncertainty indices at increased levels. Overall, results show that cryptocurrencies and traditional safe-havens (gold and U.
S. dollar index) exhibit heterogeneous hedge and safe-haven properties across market-implied and event-based uncertainties.
2. Data
We utilize two major cryptocurrency ETFs (i) S&P Bitcoin index (SPBTC) and (ii) S&P Broad digital market index (SPBDCI), along
with gold and the U.S. Dollar index (USDI). We include the CBOE volatility index (VIX), CBOE Emerging markets volatility index
(VXEEM), CBOE EFA ETF volatility index (VXEFA), CBOE crude oil volatility index (OVX), as well as the U.S. economic policy un-
certainty (EPU), and the Geopolitical risk (GPR) uncertainty indices. Asset prices and volatility data are from Bloomberg. U.S. EPU and
GPR are sourced from Baker et al. (2016) and Caldara and Iacoviello (2022) respectively.
1
Daily closing prices are matched by calendar
date for return calculations (Ali et al., 2025a).
The sample period spans from 1st January 2018, to 31st March 2025, based on data availability. All the series are converted to
returns using first-differences of the natural logarithms. Following Bouri et al. (2017), we apply first principal component analysis
(PCA) to VIX, VXEEM, VXEFA to construct a composite volatility index (CVIX). The first principal component captures shared variation
and explains approximately 44.5% of the total variance. Further details on variables description, PCA results, descriptive statistics,
correlation matrix, and BDS test are provided in Appendix (Tables A1, A2, A3, A4 and A5).
3. Methodology
3.1. GARCH model
This study follows Bouri et al. (2017) and Wu et al. (2019) to investigate the hedge and safe-haven properties of financial assets
(SPBTC, SPBDCI, gold, and USDI) against uncertainty shocks (CVIX, OVX, EPU, and GPR). We employ EGARCH (1,1) model with
dummy variables using the maximum likelihood method.
The return and variance equation (EGARCH) are specified as follows:
qt=γ+λqt−1+β0ri,t+β1D(ri,q90).ri,t+β2D(ri,q95).ri,t+β3D(ri,q99).ri,t+et(1)
ln(
σ
2
t)=θ0+θ1
et−1

ht−1
√
+θ2ln (
σ
2
t−1)+θ3
et−1

ht−1
√(2)
Here qt denotes asset returns (SPBTC/SPBDCI/gold/USDI), ri,t represent changes in uncertainty indices, where i∈(CVIX/OVX/EPU/
GPR). D(ri,qp)is a dummy variable, equal to 1 if any of uncertainty index exceeds its p-th percentile (e.g., 90 %, 95 %, 99 %), and
0 otherwise, et is the error term.
3.2. Wavelet quantile correlation (WQC)
Investors may have different preferences across distinct timescales and market states. In order to understand the cryptocurrencies,
gold and U.S. dollar index hedging and safe-haven properties across quantiles and frequencies, we propose wavelet quantile corre-
lation approach (Kumar and Padakandla, 2022), an extension based on quantile correlation proposed by Li et al. (2012).
We start the process by decomposing asset pairs X[t]and Y[t]pairs of cryptocurrencies, gold, U.S. dollar index, and uncertainty
indices by using a maximal overlap discrete wavelet transform (MODWT) (Percival and Walden, 2001).
MODWT begins by analysing the asset returns r[i]with a length T, where T is 2J for certain integer J. An orthogonal wavelet-defines
low-pass filter as g1[i]and high-pass filter as h1[i]. At first decomposition level, the convolution of r[i]with g1[i]yields approximation
coefficients a1[i], whereas its convolution with h1[i]produces the detail coefficients c1[i].Both a1[i]and c1[i]have a length denoted as
N.
1
https://www.policyuncertainty.com/ and https://www.matteoiacoviello.com/gpr.htm
P.K. Ankam and J.K. Pattanayak
Finance Research Letters 86 (2025) 108934
2
a1[i] = g1[i] ∗ j[i] = ∑
ng1[i−n]j[n]
c1[i] = h1[i] ∗ j[i] = ∑
nh1[i−n]j[n]
Thereafter, the approximation coefficient a1[i]is further decomposed using modified filters g2[i]and h2[i], obtained through dyadic
up-sampling of g1[i]and h1[i]and this process is applied recursively to obtain multilevel representations. For J=1,2…..J0−1 and
J0≤J, the approximation and detain coefficients at each level can be computed as:
aj+1[i] = gj+1[i] ∗ aj[i] = ∑
ngj+1[i−n]aj[i]
cj+1[i] = hj+1[i] ∗ aj[i] = ∑
nhj+1[l−n]aj[j]
Where gj+1[i] = U(gj[i])and hj+1[i] = U(hj[i]).
Utilizing the up-sampling function ‘U’ zero is included among every adjacent pair of time-series elements.
After applying J level decomposition to X[t]and Y[t], detail coefficients are extracted and quantile correlation is applied to each pair
of wavelet details, dj[X]and dj[Y]for all J.At a given decomposition level J and quantile (
τ
), we express WQC for X and Y time-series
as:
WQC
τ
(dj[X],dj[Y])=qcov
τ
(dj[X],dj[Y])

var(∅
τ
(dj[Y] − Q
τ
,dj[Y]))var(dj[X])
√(3)
We retrieve data at scales 2–4, 4–8, 8–16, (short-term) 16–32, 32–64, 64–128, (medium-term) 128–256, >256 (long-term) days to
capture the daily behavior. Statistical significance is assessed via a bootstrap procedure (Efron and Tibshirani, 1994) with 500 rep-
licates. The correlation coefficients that deviate the 95 % bootstrap confidence interval are considered statistically significant at 5 %
level. Although utilized in financial series data, we assumed independent and identically distributed resampling and might under-
estimate time dependence. We recommended future studies to go for subsampling or block bootstrapping methods (Politis et al., 1999)
for improved robustness.
4. Empirical results
4.1. EGARCH with dummy variables
Following Wu et al. (2019) and Das et al. (2019) based on the sign and statistical significance of β coefficients as per Eq (1), asset
returns (SPBTC/SPBDCI/gold/USDI) may act as hedge if β0>0, against uncertainty indices (CVIX/OVX/EPU/GPR) and as a
safe-haven if the cumulative coefficients satisfy ∑1
i=0βi>0 at 90 % quantile, ∑2
i=0βi>0 at 95 % quantile, and ∑3
i=0βi >0 at 99 %
quantile. If statistically significant, they are classified as strong; otherwise, weak.
Table 1 reports EGARCH estimates for the indices considered at average conditions against CVIX. SPBTC shows neither hedging
(β0<0)nor safe-haven behavior (∑k
i=0βi<0,[k=1,2,3]). Mei-Jun and Guang-Xi (2024) find increased asymmetric
cross-correlations between cryptocurrencies, G7 and E7 stock markets. SPBDCI is not a hedge but acts as safe-haven across tail dis-
tributions, particularly at 95 % and 99 % quantiles (∑k
i=0βi>0,[k=2,3]) resulting from index diversification. Gold is a strong hedge
(β0>0,significant), but lacks safe-haven properties, due to liquidity-driven sell-offs during stress (Baur and Lucey, 2010). USDI is both
a strong hedge and a weak safe-haven (∑k
i=0βi>0,[k=1,2,3]), benefitting from U.S dollar risk reduction benefits and global demand
(Sharma and Karmakar, 2022).
Table 2 shows SPBTC as a weak hedge (β0>0), without safe-haven features against OVX. SPBDCI does not act as hedge and as safe-
haven (∑k
i=0βi<0,[k=1,2,3]). Gold is a weak hedge (β0>0), while USDI acts as strong hedge and as weak safe-haven at 90 %
(∑1
i=0βi>0), and 99 % quantiles (∑3
i=0βi>0). The results coincide with Das et al. (2019) who revealed dollar as superior to Bitcoin
for hedging oil volatility.
Against EPU, Table 3 shows SPBTC acts as weak safe-haven at extreme market conditions as shown at 90 % quantile(∑1
i=0βi>0).
SPBDCI shows weak safe-haven roles at 95 % and 99 % quantiles. Gold exhibits weak hedging and weak safe-haven behavior, while
USDI is not a hedge, but fulfills weak safe-haven role at 95 % quantile (∑2
i=0βi>0).These results reinforce the findings of Wu et al.
(2019) regarding the limited safe-haven roles of Bitcoin and gold under policy related uncertainty.
P.K. Ankam and J.K. Pattanayak
Finance Research Letters 86 (2025) 108934
3
Table 1
Regression results.
Parameters γ λ β0∑1
i=0βi∑2
i=0βi∑3
i=0βiθ0θ1θ2θ3
SPBTC 0.000862* 0.011581 −0.003751*** −0.0072003* −0.003221 −0.004806 −0.672168*** −0.018068 0.893211*** 0.205809***
(0.000464) (0.015622) (0.000548) (0.000737) (0.000889) (0.002088) (0.141427) (0.014857) (0.022115) (0.032250)
SPBDCI 0.000064 0.035570 −0.002369 −0.005399 0.000973 0.002534 −0.563466 −0.025903 0.909292*** 0.186767***
(0.000981) (0.025376) (0.001186) (0.002780) (0.001802) (0.002083) (0.254625) (0.017119) (0.040446) (0.046199)
Gold 0.000585 0.002430 0.000547** −0.000590 −0.000005 −0.000908 −0.298714*** 0.054347*** 0.968000*** 0.139883***
(0.000235) (0.024679) (0.000286) (0.001208) (0.000424) (0.000391) (0.021537) (0.012895) (0.002150) (0.016354)
USDI 0.000076 0.024237 0.000451*** 0.001059 0.000267 0.000001 −0.233037*** −0.009485 0.978643*** 0.133421***
(0.000113) (0.027088) (0.000100) (0.00026) (0.000159) (0.000170) (0.011563) (0.012452) (0.000994) (0.020547)
Note: Hedge and safe-haven results of SPBTC, SPBDCI, Gold, USDI against CVIX. If β0>0, specific asset is hedge against CVIX. ∑k
i=0βi>0(k=1,2,3)denotes specific assets act as safe haven against
CVIX for 90 %, 95 % and 99 % quantiles, respectively.
γ =constant term, λ =lagged dependent variable. θ0,θ1,θ2,θ3 =Base volatility, Shock impact, volatility persistence, and leverage effect.
***, **, and * denote 1 %, 5 % and 10 % significant levels, respectively. Each estimate is followed by its standard error in parentheses.
P.K. Ankam and J.K. Pattanayak
Finance Research Letters 86 (2025) 108934
4
Table 2
Regression results.
Parameters γ λ β0∑1
i=0βi∑2
i=0βi∑3
i=0βiθ0θ1θ2θ3
SPBTC 0.001523** 0.004021 0.029102 −0.019480 −0.062756 −0.04425 −0.772151*** −0.016182 0.877292*** 0.224769***
(0.000655) (0.018867) (0.017426) (0.00461) (0.01661) (0.026208) (0.162063) (0.015879) (0.025362) (0.033903)
SPBDCI 0.001701** 0.033470** −0.017814 −0.11874 −0.135288 −0.093303 −0.527654 −0.015447 0.915063*** 0.186567*
(0.000786) (0.016993) (0.021147) (0.038405) (0.035393) (0.026852) (0.555675) (0.033055) (0.088261) (0.097593)
Gold 0.000469** 0.004614 0.000184 −0.006523 −0.006300 −0.00540 −0.309254*** 0.050203*** 0.966847*** 0.143099***
(0.000157) (0.023829) (0.004265) (0.010466) (0.006946) (0.006099) (0.029471) (0.012835) (0.002921) (0.020534)
USDI 0.000088* 0.016747 0.001291* 0.002849 −0.001969 0.001662 −0.211456*** −0.001749 0.980554*** 0.128930***
(0.000052) (0.016600) (0.000768) (0.00109) (0.001441) (0.001648) (0.006912) (0.011815) (0.000594) (0.012134)
Note: Hedge and safe-haven results of SPBTC, SPBDCI, Gold, USDI against OVX.
***, **, and * denote 1 %, 5 % and 10 % significant levels, respectively. Each estimate is followed by its standard error in parentheses.
P.K. Ankam and J.K. Pattanayak
Finance Research Letters 86 (2025) 108934
5