The impact and profitability of day trading following the relaxation of day trading restrictions in Taiwan

Abstract

The relaxation of day trading restrictions in Taiwan at the start of 2016 resulted in a significant increase in day trading volume, which piqued our interest in researching the impact and profitability of day trading, expected (unexpected) day trading, and day trading at high (low) levels of VIX using time series models, with the following key findings. First, we show that a high market trading volume triggers a high day trading volume resulting from liquidity markets that day traders prefer, but a high day trading volume does not trigger a high market trading volume resulting from speculative markets that other market participants don't prefer. Second, contrary to our perception, while the VIX index rises, day trading would be more profitable after the relaxation. We infer that a high VIX index may be accompanied by a volatile market, which may generate profits by widening the intraday spread of a day-tradable stock. Third, as compared with unexpected market trading volume, we reveal that unexpected day trading volume may be more unexpected than market trading volume, being more likely to enhance market volatility and stock returns. These impressive and interesting findings may not be disclosed before the relaxation, which may contribute to the existing literature.

1. Introduction

Based on the efficient market hypothesis (EMH), stock prices might not be difficult to predict since all available information is fully reflected in stock prices [1,2]. In other words, the information released by day trading relaxation might not impact stock price based on EMH. However, Schwert [3] contends that anomalies appear inconsistent with the EMH, implying that market inefficiency may still exist in stock markets [100]. For example, disposition effects [4], overreaction hypothesis [5], and herding behaviors [6] challenge EHM. Furthermore, if a market is efficient, there is no lead-lag relationship between markets [7,8].

However, the relationships between market trading volume, trading volatility, and return are widely explored in relevant studies [[9], [10], [11], [12], [13], [14], [15]]. Because of the relaxation of the day trading restriction (hereafter referred to as the relaxation) at the beginning of 2016 in Taiwan, this study shows that day trading volume has increased considerably on the Taiwan Stock Exchange (TWSE). The above phenomenon arouses our interest in investigating the profitability in day trading, expected (unexpected) day trading, and day trading at the high (low) levels of VIX. Furthermore, we argue that our investigated issues from multiple views would be the study's novelty and significance because our issues relate not only to shocks (e.g., unexpected daytrading volume) but also to human nature (e.g., greed and panic gauged by the VIX index). As a result, we claim that this study explores the impact and profitability of day trading in Taiwan from a variety of perspectives that are rarely addressed in the existing literature.

We state that the trading behaviors of day traders may be unlike those of other investors (e.g., long-term investors), which may be due to either investing strategies employed by day traders [16] or the characteristics of day trading [17]. For example, since day traders unnecessarily hold stocks overnight, they might not necessarily have abundant capital and even suffer losses if bad news occurs after the market closes.

Therefore, the gain (loss) of day trading stock would be determined by the difference between the buy price and sell price of a stock after taking transaction cost into account. As such, day traders may prefer trading stocks with high volatilities (e.g., high-beta stocks) and stocks without liquidity risk [18,19]. Additionally, if the relaxation would enhance market trading volume, the liquidly risk of stock markets would be decreased, thereby likely appealing to more investors trading in stock markets. Thus, after the relaxation, we argue that more investors may day trade stocks and even generate profits (suffer losses) if these investors employ appropriate (inappropriate) investing strategies.

Based on previous studies [20,21,99], day traders would be attributed as losers. However, Kuo and Lin [22] argue that day traders with unreleased information may make profits, indicating that some of the day traders might not be regarded as losers rather than winners if these day traders have insider information and released information [16,23]. Because of the mixed results discussed above, we conclude that the profitability of day trading after the relaxation is worth further investigation due to the importance of this study in light of the literature. Moreover, we state that even though day trading volume may contribute to the daily market trading volume after the relaxation, the issues related to day trading seem seldom explored compared with trading activities in general.

In addition, since day traders are deemed as uninformed traders, they might induce informed traders to manipulate share prices for these uninformed day traders, thereby enhancing the trading performance of these informed traders [24,25]. However, if day trading is greatly prevalent, it might frighten long-term investors outside the market because the market is likely regarded as an excessive speculative market. As such, we argue that the lead-lag relationship between market trading activities and day trading activities would be worthy of further investigation since examining the lead-lag relationship between trading volume and returns is an important issue in relevant studies [9,26,27].

Apart from the lead-lag relationship [[28], [29], [30], [31]], the contemporary association between stock return, trading volume, and trading volatility would be examined in previous studies [32,33] since volume may affect return through the channel of volatility [34,35]. Because of the concern about whether day trading volume and market trading volume would have different impacts on the stock market, we then explore whether different trading volumes (i.e., day and market trading volumes) would have diverse impacts on volatility, thereby differently affecting return. In addition, we also classify trading volume into expected and unexpected trading volumes since unexpected trading volume is likely regarded as unreleased information that may strongly affect return as well [[35], [36], [37]] Moreover, Wang, Chih, and Chou [38] state that individual investors contribute more than 50% market trading volume of the TWSE, resulting in the TWSE that is vulnerable to investor sentiment. As such, we further explore if the profitability of day trading would be affected by either the “fear” gauged by a rather high level of VIX or the “greed” measured by a fairly low level of VIX.

Based on the above, we are interested in the profitability of day trading, expected (unexpected) day trading, and day trading during the high (low) level of VIX in this study, and then reveal the following impressive findings that may contribute to the existing literature. First, we show that a high market trading volume would trigger a high day trading volume, which might result from liquidity markets that are preferred by day traders, but a high day trading volume would not initiate a high market trading volume, which might arise from speculative markets that are not preferred by other market participants. Second, different from our perception, we disclose that, while the VIX index rises at a rather high level, the profitability of day trading would be enhanced after the relaxation. We infer that a higher VIX index may be accompanied by a highly volatile market, thereby likely generating profits by widening the intraday spread of a stock that would be advantageous for day trading. Third, we also disclose that compared with unexpected market trading volume, unexpected day trading volume would have more impact on stock returns via the channel of stock volatility. We deduce that unexpected day trading volume might be more unexpected than unexpected market trading volume, thus intensifying stock market volatility and further enhancing stock returns.

2. Literature review and hypotheses proposed

2.1. Informational trading activities

The following trading variables can be used to assess trading activity: trading quantity, trading frequency, and trading size [39]. According to O'Neill and Swisher [40], the total trades and trade size both convey information to investors, and it has been found that informed traders favor trading stocks with higher trading volumes. Kim and Verrecchia [41] emphasize the importance of trading size in terms of information content. In addition, relevant research has shown that transaction costs may account for the observed trade volume [42,43], volume sequences can be informative [44,45], and the historical trading volume serves to reconcile underreaction and overreaction in stock markets, rather than serving as a significant link between momentum and value strategies [46,47].

2.2. The relationship between trading activities and stock markets

Some investors try to incorporate released and even unreleased information into their trading activity in order to predict future returns and volatility [48,49]. Relevant research has explored the impact of trading volume on volatility [50,51], as evidenced by the fact that market volatility and trading activities are related [52,53]. While exploring the association between trading volume and volatility in financial markets, unexpected volume shocks significant impact on volatility [54,55], positive unexpected volume shocks greatly affect volatility as compared with negative unexpected volume shocks [56], and the importance of trade size in the volatility–volume relationship of stock markets is confirmed [57,58]. Moreover, Xu, Chen, and Wu [12] demonstrate that volatility and volume are durable and have a strong correlation with historical volatility and volume. Dungore et al. [59] also reveal that the more day trading, the higher volatility. Koubaa and Slim [57] further show that information released in stock markets significantly affects trading volume and volatility, confirming the volume-volatility relationship.

Nishimura and Sun [60] also looked at the impacts of intraday trading volume on trading volatility in the futures markets of China by using intraday data. They reveal that trading volume positively impacts volatility in such markets. Furthermore, Darrat, Zhong, and Cheng [61] show the interrelationship between trading volume and volatility while public information is released. Nonetheless, while no public information is released, they show that trading volume on trading day t-1 affects volatility on trading day t without feedback. Furthermore, while relevant research has indicated a lead-lag link between return and volume, the lead-lag relationship between day trading volume and market trading volume appears to be understudied in the relevant studies. Moreover, it would be worthwhile to investigate whether the lead-lag relationship between return and different trading activity (market trading volume vs. day trading volume) differs. Then we proposed H1-1 and H1-2.

H1-1: There is a bi-directional relationship between market trading volume and day trading volume.

H1-2: There is a bi-directional relationship between trading activities and stock return.

2.3. The relationship between return, volume, and volatility

Previous studies have found bidirectional causality between daily returns and volume [14,15]. Stocks with abnormally high trading volume for more than a week are more likely to appreciate in subsequent periods [62,63]. Concerning the relationships between volume, volatility, and return, Hsieh, Chen, and Hwang [64] state that, based on trading volatility regarded as the temporal reaction of returns, trade volume, and the number of trades, high-volume trading activity may be attributed to the increasing volatility in either returns or the number of trade. In addition, Gurgul and Syrek [65] investigate the interrelationship among returns, volatility, and volumes for the constituent stocks of either CAC 40 or FTSE 100 and find the interrelationship between volume and volatility for these constituent stocks. Heston, Korajczyk, and Sadka [66] show that volume and volatility may not adequately explain return predictability.

Apart from exploring the lead-lag relationship by the VAR model, this study would further explore the contemporary relationship between return, volume, and volatility. After surveying the relevant studies [35,[67], [68], [69]], this study may adopt the EGARCH-in mean model (i.e., EGARCH(1,1)-M model) to examine whether stock return would be impacted by volume through the channel of conditional volatility. Furthermore, because we argue that daily trading volume is less likely to be predicted than market trading volume, we infer that the information on day trading volume seems to be stronger than that on market, which may result in stock returns being more affected by day trading volume than market trading volume via conditional volatility. As a result, we proposed H2.

H2

The information on day trading volume would have a greater impact on stock return than market trading volume.

Aside from exploring the lead-lag relationship and contemporary relationship in the above hypotheses, we argue that there is still room for further investigation. Todorova and Clements [39] discover that volume is slightly more informative than volatility and that both expected and unexpected trading volume appears to be strongly related to volatility when studying the association between volume and volatility. By employing ARIMA models, we thus retrieve the expected and unexpected day (market) trading volume and then examine whether either the unexpected or expected day (market) trading volume would matter to stock return, which may provide more information in terms of day trading volume on stock markets.

Similar to the inference for H2, we deduce that the information on unexpected trading volume would have more impact on the market, resulting in stock returns being more impacted by unexpected day (market) trading volume than expected day (market) trading volume via conditional volatility. As a result, we proposed H3.

H3

Unexpected trading volume would have more impact on stock return compared with expected trading volume.

In addition, the VIX index, an important indicator of measuring market sentiment, is highly negatively correlated with the trend of the S&P 500 index [[70], [71], [72]]. Moreover, if the VIX is high, it indicates that the market might be full of panic sentiment [71,101], and if the VIX index remains low, it also indicates that the market might be full of optimistic sentiment [12]. Based on that the VIX index is a market implied volatility [70], we thus further examine whether stock return could be impacted by day trading volume through conditional volatility. However, the threshold point of either the high VIX or low VIX index might not have a unanimous criterion. We thus employ the wisdom of 0.9 and 0.1 quantile commonly adopted in quantile regression as the threshold point of either high VIX or low VIX index. In other words, if the VIX index reaches its highs, it can cause the stock market to fall to a lower level, representing that the stock market may rise later [[73], [74], [75], [76]]. As a result, we proposed H4 to incorporate the VIX index into the EGARCH(1,1)-M model. Furthermore, as far as we know, the aforementioned issue may not have been addressed in the existing literature.

H4

Day trading volume would affect stock returns through the channel of conditional volatility by incorporating the interaction term (i.e., volume multiplied by the VIX index at either 90% decimal or 10% decimal) into the conditional volatility.

2.4. Further investigation

Based on the fact that trading volume increased significantly after the relaxation, we investigate whether the stock market will be affected differently before and after the relaxation. As a result, we offer Hypothesis 5.

H5

The impact of day trading volume on the stock market would differ before and after the relaxation.

3. Methodology

3.1. The lead-lag relationship between return and trading activities

Although the association between return and volume is widely explored in previous studies [[77], [78], [79], [80]], Chung et al. [81] also reveal that not only day trading activity would positively affect return volatility but also past volatility would positively affect future day trading activity. As such, we argue that the above phenomena would be worthy of further investigation before and after the relaxation in Taiwan.

By exploring the interrelationship between market trading volume and day trading volume before and after the relaxation, we then employ the VAR model1 (i.e., Model (1)) to test Hypotheses 1-1 – 1-2.

$${\mathrm{y}}_t={\left[R_t,{MV}_t,{DT}_t\right]}^{\prime }=A_0+{\mathrm{\Sigma }}_{i=1}^p{B}_i{y}_{t-i}+{\epsilon }_t$$ (1)

where ${\mathrm{y}}_t$ is a vector of three endogenous variables, including stock market return ($R_t$), the change rates of market trading volume (${MV}_t$), and day trading volume (${DT}_t)$. In addition, the variables including $R_t,{MV}_t,{\mathrm{a}\mathrm{n}\mathrm{d}DT}_t$ are all log-difference time series. The matrix $B_i$ is estimated and the coefficients of this matrix are $b_{Ri}$, $b_{MVi}$, and $b_{DTi}$ for stock return, market trading value, and day trading volume, respectively. ${\epsilon }_t$ is a vector of innovations.

3.2. The dynamic effect of day trading volume

Apart from exploring the lead-lag relationship by the VAR model, we would further explore the contemporary relationship between return, volume, and volatility. After surveying the relevant studies [35,[67], [68], [69]], we adopt the EGARCH-in mean model (i.e., EGARCH(1,1)-M model) to examine whether stock return would be impacted by volume through the channel of conditional volatility. We thus test Hypothesis 2 by using Model (2).

$${\mathrm{R}}_t={\beta }_0+{{\mathrm{\Sigma }}_{i=1}^2\beta }_{Ri}{R}_{t-i}+{\beta }_h\log h_t+u_t,u_t\sim N\left(0,h_t\right)$$

$$\log h_t=c_0+a\left|u_{t-1}\right|/\sqrt{h_{t-1}}+b\log h_{t-1}+d{u}_{t-1}/\sqrt{h_{t-1}}+\omega {Volume}_t+{\nu }_t$$ (2b)

where $R_t$ is the log-return of the stock index, $h_t$ is conditional volatility, $\log h_t$ includes the terms proposed by Nelson [69], ${Volume}_t$ is market trading volume or day trading volume.

3.2.1. Concerning unexpected trading volume

In addition to exploring the effect of unexpected trading activity on stock markets [57,82,83], we state that either unexpected day trading volume at times t and t-1 [84] or even positive or negative unexpected day trading volume may have different impacts on conditional volatility [77]. Furthermore, increasing unexpected trading volumes may have higher volatility than decreasing unexpected trading volumes since there is a significant positive correlation between volatility and unexpected volumes [56]. We thus test H3 based on Model (3).

$${\mathrm{R}}_t={\beta }_0+{{\mathrm{\Sigma }}_{i=1}^2\beta }_{Ri}{R}_{t-i}+{\beta }_h\log h_t+u_t,u_t\sim N\left(0,h_t\right)$$

$$\log h_t=c_0+a\left|u_{t-1}\right|/\sqrt{h_{t-1}}+b\log h_{t-1}+d{u}_{t-1}/\sqrt{h_{t-1}}$$

$$+{\omega }_e{Volume}_t^e+{\omega }_{u0}{Volume}_t^{un}+{\omega }_{u1}{Volume}_{t-1}^{un}+{\omega }_{n}D\bullet {Volume}_t^{un}+{\nu }_t$$ (3b)

where ${Volume}_t^e$ is the expected volume at t, ${Volume}_t^{un}$ is the unexpected volume at t, ${Volume}_{t-1}^{un}$ is the unexpected volume at t-1, and D is a dummy variable. D is set to 1 if the unexpected volume is positive; otherwise, set to 0.

3.2.2. Concerning the VIX index

Consequently, by incorporating the day trading volume that interacts with the VIX index at 90% decimal or 10% decimal through the channel of the conditional volatility in Model (4), we then examine Hypothesis 4.

$${\mathrm{R}}_t={\beta }_0+{{\mathrm{\Sigma }}_{i=1}^2\beta }_{Ri}{R}_{t-i}+{\beta }_h\log h_t+u_t,u_t\sim N\left(0,h_t\right)$$

$$\log h_t=c_0+a\left|u_{t-1}\right|/\sqrt{h_{t-1}}+b\log h_{t-1}+d{u}_{t-1}/\sqrt{h_{t-1}}$$

$$+\omega {Volume}_t+{\omega }_H{Volume}_t\bullet D_{VIXH,t}+{\omega }_L{Volume}_t\bullet D_{VIXL,t}+{\nu }_t$$ (4b)

where $D_{VIXH,t}$ is a dummy variable as VIX falls into over 0.9 quantiles, and $D_{VIXL,t}$ is a dummy variable as VIX falls into below 0.1 quantiles.

4. Data and empirical results

4.1. Descriptive statistics

We collect daily data for day trading volume from the web of the Taiwan Stock Exchange due to the data only being released on this web. Table 1 reports the means, standard errors, maxima, and minima of the Taiwan stock index, market trading volume, and day trading volume over the period 2014–2017.

Table 1.

Descriptive statistics.

	Raw data			Change rate (%)
	Stock index	Market trading volume (million)	Day trading volume (million)	Stock index	Market trading volume	Day trading volume
Full sample period
Mean	9233.0325	91915.6	9266.99	0.0229	0.0137	0.3815
Std. error	730.8894	20858.6	7991.80	0.7832	17.5615	19.7303
Min..	7410.3400	30206.9	663.39	−4.9569	−76.3383	−83.7303
Max..	10854.5700	233551.0	48682.30	3.5175	74.3869	101.6313
Period 1
Mean	8931.9780	92167.3	5137.56	−0.0083	−0.0843	0.4711
Std. error	517.1191	16765.1	2670.31	0.8783	16.2601	20.4613
Min..	7410.3400	39334.8	663.39	−4.9569	−59.6366	−72.1295
Max..	9973.1200	168637.0	16535.90	3.5175	74.3869	101.6313
Period 2
Mean	9561.0471	91641.4	13766.20	0.0570	0.1204	0.2841
Std. error	786.9471	24570.2	9323.63	0.6638	18.8921	18.9246
Min..	8053.6900	30206.9	3269.34	−3.0202	−76.3383	−83.7303
Max..	10854.5700	233551.0	48682.30	2.7758	72.8214	99.3656

Notes: There are 980 samples over the data period (i.e., from Jan. 6, 2014, to Dec. 29, 2017). Due to day trading expanding considerably after the relaxation on Feb 1, 2016, we then separate two periods (i.e., Period 1 includes 511 samples before the relaxation, and period 2 includes 469 samples after the relaxation).

Fig. 1 displays that the downward trend of the Taiwan stock index is shown before the relaxation (i.e., period 1: from Jan. 6, 2014, to Jan. 31, 2016), but the upward trend of the Taiwan stock index appears after the relaxation (i.e., period 2: from Feb. 1, 2016, to Dec. 29, 2017); period 1 includes 511 samples before the relaxation, and period 2 includes 469 samples after the relaxation. Regarding market trading volume, we reveal that the volatilities of market trading volume are rather higher in period 2 instead of period 1 as shown the standard errors are 24570.2 in period 2 compared with 16765.1 in period 1, even though the mean market trading volume in period 1 is slightly higher than that in period 2, as shown that the mean values are 92167.3 and 91641.4 for period 1 and period 2 in Table 1.

Fig. 1.

The Movement in levels of the stock index, day trading volume, and market trading volume. Note: The stock index is monitored on the right vertical axis and the market or day trading volume is on the left. Feb. 1, 2016, is marked by the vertical solid line in this picture.

Regarding day trading volume, we observe that day trading volume significantly increases in period 2 as revealed that the day trading volume in period 2 is about 2.7 times of day trading volume in period 1, indicating that day trading volume considerably increases in period 2.

4.2. The lead-lag relationship between return and trading activities

We explore the association between stock index return, market trading volume, and day trading volume by employing the VAR model [85] in Model (1). According to the lag-length test, we choose lag 2 for this VAR model and the results are presented in Table 2.

Table 2.

The result of the VAR model in terms of $R_t,{MV}_t,{andDT}_t$. ${\mathrm{y}}_t={\left[R_t,{MV}_t,{DT}_t\right]}^{\prime }=A_0+{\mathrm{\Sigma }}_{i=1}^p{B}_i{y}_{t-i}+{\epsilon }_t$ (1).

	$R_t$	${MV}_t$	${DT}_t$
$R_{t-1}$	0.0355	0.4328	−2.4045***
	(1.1060)	(0.6507)	(-3.2897)
$R_{t-2}$	−0.0142	0.2578	−0.7840
	(-0.4358)	(0.3823)	(-1.0578)
${MV}_{t-1}$	0.0026	−0.3492***	0.1450***
	(1.1743)	(-0.6711)	(2.8982)
${MV}_{t-2}$	−0.0003	−0.0730	0.1441***
	(-0.1541)	(-1.6151)	(2.8996)
${DT}_{t-1}$	−0.0001	−0.0580	−0.5084***
	(-0.0383)	(-1.4240)	(-11.3478)
${DT}_{t-2}$	−0.0001	−0.1334***	−0.3296***
	(-0.0272)	(-3.3313)	(-7.4913)
Constant	0.0219	0.0537	0.6855
	(0.8698)	(0.1032)	(1.1983)
F statistic
$H_o:{\mathrm{\Sigma }}_{i=1}^2{b}_{Ri}=0$	0.6887	0.2950	6.1163***
	(0.5025)	(0.7446)	(0.0023)
$H_o:{\mathrm{\Sigma }}_{i=1}^2{b}_{MVi}=0$	0.8939	30.3342***	6.1172***
	(0.4094)	(0.0000)	(0.0023)
$H_o:{\mathrm{\Sigma }}_{i=1}^2{b}_{DTi}=0$	0.0008	5.5701***	70.6735***
	(0.9992)	(0.0039)	(0.0000)

Notes: $R_t$ is the stock market return at time t, ${MV}_t$ is the change rates of market trading volume at time t, and ${DT}_t$ is the change rates of day trading volume at time t. $b_{Ri}$, $b_{MVi}$, and $b_{DTi}$ are coefficients for stock return, market trading volume, and day trading volume. The t values are shown in parentheses in the upper part of Table 2. Granger causality F statistics and their p values shown in parentheses are presented in the lower part of Table 2. Statistical significance is set at 10%, 5%, and 1% levels denoted by *, **, and ***.

Table 2 shows that day trading volume at time t would be enhanced (reduced) if market trading volume increases (decreases) at time t-1. We infer that market trading volume decreases at time t-1 would reduce day trading volume at time t because day traders might not able to complete day trading if the market with serious liquidity issues (i.e. lower market trading volume). However, increasing day trading volume at lag 1 may reduce the market trading volume. We infer that a higher day trading volume might be regarded as a signal of excessive speculation, which might hinder long-term investors in trading stocks, thereby, declining market trading volume. Therefore, Hypothesis 1-1 is supported (i.e., F statistics (i.e., 5.5701 and 6.1172) are significant).

Moreover, we show that day trading volume is relatively sensitive to previous stock returns as shown that day trading volume at time t would be enhanced significantly after declining share prices at period t-1 as the coefficient (−2.4045) is statistical significance at 1% level in Table 2. In other words, day traders might prefer trading stocks at time t while share prices fall at time t-1. We infer that lower prices might appeal to day traders to enter the market, thereby enhancing day trading volume. However, day trading volume and market trading volume at time t-1 might not affect share prices at time t. Thus, Hypotheses 1–2 are not supported (i.e., F statistics (i.e., 0.8939 and 0.0008) are insignificant).

4.3. The dynamic effect of day trading volume

We further examine the impact of day trading on the stock market by concerning the volatility factor. We thus incorporate trading volume into the conditional volatility equation of the model (i.e., EGARCH-in-mean model), implying that trading volume may affect stock returns through the channel of conditional volatility. Table 3 also reveals that trading volume including market trading volume and day trading volume would affect conditional volatility as the coefficients of ${Volume}_t$ (0.0312 and 0.0104) are all significant at a 1% level, representing that increasing trading volume would enhance stock volatility. In addition, we also reveal that stock returns are more affected by day trading volume than market trading volume through conditional volatility, supporting Hypothesis 2. We deduce that day trading volume is more unpredictable than market trading volume and is thus more likely to enhance stock returns.

$${\mathrm{R}}_t={\beta }_0+{{\mathrm{\Sigma }}_{i=1}^2\beta }_{Ri}{R}_{t-i}+{\beta }_h\log h_t+u_t,u_t\sim N\left(0,h_t\right)$$

$$\log h_t=c_0+a\left|u_{t-1}\right|/\sqrt{h_{t-1}}+b\log h_{t-1}+d{u}_{t-1}/\sqrt{h_{t-1}}+\omega {Volume}_t+{\nu }_t$$ (2)

Table 3.

The result of the EGARCH(1,1)-M model concerning volume.

	Market trading volume		Day trading volume
	coefficient	std. error	Coefficient	std. error
${\beta }_0$	0.0302	0.0342	0.1089**	0.0492
${\beta }_{R1}$	0.0108	0.0289	0.0288	0.0333
${\beta }_{R2}$	−0.0511*	0.0298	−0.0031	0.0309
${\beta }_h$	0.0184	0.0224	0.1127***	0.0391
$c_0$	−0.1059**	0.0446	−0.0694***	0.0207
$a$	0.0907**	0.0431	0.0579***	0.0216
$b$	0.9589***	0.0164	0.9701***	0.0105
$d$	−0.1580***	0.0233	−0.1047***	0.0165
$\omega$	0.0312***	0.0026	0.0104***	0.0020
Log-likelihood function	−984.7956		−1050.1550
Model diagnosis
Ljung-Box Q(10)	9.0753		8.3415
McLeod-Li(10)	7.6839		7.3599

Note: The results using market trading volume and day trading volume are shown in Column (1) and Column (2). Ljung-Box and McLeod-Li are standard tests for autocorrelation in the levels and squares. These two statistics are asymptotic chi-squared distributions with the degrees of freedom shown in parentheses. Statistical significance is set at 10%, 5%, and 1% levels denoted by *, **, and ***.

Furthermore, the above findings arouse our interest in further exploring whether unexpected and expected day (market) trading volume would have different impacts on stock markets. We thus decompose trading volume into expected and unexpected trading volume by adopting ARMA models as shown in Model (5).

$${Volume}_t=a_0+{\mathrm{\Sigma }}_{p=1}^m{a}_p{Volume}_{t-p}+{\epsilon }_t+{\mathrm{\Sigma }}_{q=1}^n{b}_q{\epsilon }_{t-q}$$ (5)

where ${Volume}_t$ represents either market trading volume (${MV}_t$) or day trading volume (${DT}_t$) employed in this study. ${\epsilon }_t$ is the residual of the above equation. According to the Akaike Information Criterion (AIC), we select ARMA (5,1) for market trading volume and ARMA (3,5) for day trading volume. The estimated value of ARMA models is defined as expected volume (${Volume}_t^e$) and unexpected volume (${Volume}_t^{un}$) is defined as actual volume minus expected volume. In addition, since market volatility might be greatly influenced by increasing unexpected trading volume instead of decreasing unexpected trading volume, D is set to 1 in Model (3) if the unexpected volume is positive; otherwise, set to 0.

Besides, we also incorporate ${Volume}_{t-1}^{un}$ in Model (3) since the news related to unexpected volume might not react completely at time t-1 resulting in a subsequent effect likely existing at time t. As noted by Chen and Tai [35], unexpected components employed as the proxies for new information released might have more impacts on financial markets compared with expected components. We thus set up Hypotheses 3 proposed above.

Table 4 reveals that expected trading volume insignificantly affects conditional volatility, while the unexpected day (market) trading volumes significantly affect conditional volatility, supporting Hypothesis 3. In addition, similar to the results revealed in Table 3, Table 4 also shows that the market (day) trading volume insignificantly (significantly) impacts stock returns through conditional volatility as the coefficient of ${\beta }_h$, 0.0155 (0.1199), is insignificant (statistically) in using market (day) trading volume in EARCH-M model (i.e., Model 3), indicating that day trading volume instead of market trading volume does matter for stock markets through the channel of conditional volatility.

$${\mathrm{R}}_t={\beta }_0+{{\mathrm{\Sigma }}_{i=1}^2\beta }_{Ri}{R}_{t-i}+{\beta }_h\log h_t+u_t,u_t\sim N\left(0,h_t\right)\log h_t=c_0+a\left|u_{t-1}\right|/\sqrt{h_{t-1}}+b\log h_{t-1}+d{u}_{t-1}/\sqrt{h_{t-1}}+{\omega }_e{Volume}_t^e+{\omega }_{u0}{Volume}_t^{un}+{\omega }_{u1}{Volume}_{t-1}^{un}+{\omega }_{n}D\times {Volume}_t^{un}+{\nu }_t$$ (3)

Table 4.

The result of the EGARCH(1,1)-M model with expected and unexpected volume.

	Market trading volume		Day trading volume
	Coefficient	std. error	Coefficient	std. error
${\beta }_0$	0.0293	0.0393	0.1137**	0.0482
${\beta }_{R1}$	−0.0021	0.0256	0.0193	0.0304
${\beta }_{R2}$	−0.0567*	0.0293	−0.0092	0.0290
${\beta }_h$	0.0155	0.0243	0.1199***	0.0450
$c_0$	−0.1042***	0.0376	−0.0545***	0.0227
$a$	0.0786***	0.0300	0.0307	0.0197
$b$	0.9698***	0.0107	0.9727***	0.0080
$d$	−0.1278***	0.0204	−0.0963***	0.0143
${\omega }_e$	0.0094	0.0063	0.0031	0.0027
${\omega }_{u0}$	0.0364***	0.0028	0.0160***	0.0022
${\omega }_{u1}$	−0.0015	0.0039	0.0021	0.0025
${\omega }_n$	0.0019	0.0017	0.0011	0.0008
Log-likelihood function	−966.9749		−1039.1409
Model diagnosis
Ljung-Box Q(10)	9.0753		8.9093
McLeod-Li(10)	7.6839		8.5588

Note: The results using market trading volume and day trading volume are shown in Column (1) and Column (2). Volume variables are divided into expected and unexpected components by employing ARMA(p, q) models.

By incorporating VIX into our models, our revealed results would not support Hypothesis 4 since both coefficients $({\omega }_H$ and ${\omega }_L$) are not significant in Table 5. Thus, contrary to our prediction, neither high decimal VIX nor lower decimal VIX significantly affects stock volatility. However, day trading volume significantly positively affects stock returns through conditional volatility. In addition, based on using day trading volume of diverse EGRACH-M models (i.e., Models (2)–(4)), Table 3, Table 4, Table 5 show that both coefficients ($\omega$ and ${\beta }_h$) are statistically significant, indicating that day trading volume instead of market trading volume would affect stock returns through the channels of conditional volatility.

$${\mathrm{R}}_t={\beta }_0+{{\mathrm{\Sigma }}_{i=1}^2\beta }_{Ri}{R}_{t-i}+{\beta }_h\log h_t+u_t,u_t\sim N\left(0,h_t\right)$$

$$\log h_t=c_0+a\left|u_{t-1}\right|/\sqrt{h_{t-1}}+b\log h_{t-1}+d{u}_{t-1}/\sqrt{h_{t-1}}+\omega {Volume}_t+{\omega }_H{Volume}_t\times D_{VIXH,t}+{\omega }_L{Volume}_t\times D_{VIXL,t}+{\nu }_t$$ (4)

Table 5.

The result of the EGARCH(1,1)-M model with the interaction in terms of VIX.

	Volume: Market trading volume		Volume: Day trading volume
	coefficient	std. error	coefficient	std. error
${\beta }_0$	0.0335	0.0398	0.1209**	0.0474
${\beta }_{R1}$	0.0084	0.0313	0.0310	0.0315
${\beta }_{R2}$	−0.0526*	0.0289	−0.0012	0.0327
${\mathit{\beta }}_{\mathit{h}}$	0.0207	0.0274	0.1275***	0.0481
$c_0$	−0.1231**	0.0501	−0.0713***	0.0181
$a$	0.1042**	0.0461	0.0594***	0.0194
$b$	0.9502***	0.0213	0.9702***	0.0099
$d$	−0.1600***	0.0237	−0.1092***	0.0173
$\omega$	0.0317***	0.0028	0.0100***	0.0019
${\mathit{\omega }}_{\mathit{H}}$	0.0009	0.0047	0.0004	0.0030
${\mathit{\omega }}_{\mathit{L}}$	−0.0062	0.0063	0.0034	0.0033
Log-likelihood function	−984.2159		−1049.6197
Model diagnosis
Ljung-Box Q(10)	8.8955		8.4679
McLeod-Li(10)	8.3687		7.7121

Note: The results using market trading volume and day trading volume are shown in Column (1) and Column (2).

5. Further investigation

After 2016, day trading volume increased significantly. we then investigate two sub-periods including Period 1 (i.e. Jan. 2014–Jan. 2016), and Period 2 (i.e. Feb. 2016–Dec. 2017) for investigating whether day trading volume would impact stock markets through the channel of conditional volatility. Based on the concern of whether the results would be different between Period 1 and Period 2, we thus examine the role of day trading volume before and after the relaxation by following the process in Section 4.

5.1. The lead-lag relationship in sub-periods

While comparing the VAR model results of Period 1 (Table 6) to those of Period 2 (Table 7), this study reveals that the role of day trading would be different after the relaxation. First of all, the results in Period 2 support Hypothesis 1-1, indicating that the day trading volume plays an important role after the relaxation. However, while we examine the relationship between day trading volume and market trading volume, we revealed that day trading volume at time t-1 would negatively affect market trading volume. We infer that market participants might regard the markets with higher day trading volume as speculative markets, thereby reducing market trading volume.

Table 6.

The result of the VAR model for Period 1 (Jan. 2014–Jan. 2016). ${\mathrm{y}}_t={\left[R_t,{MV}_t,{DT}_t\right]}^{\prime }=A_0+{\mathrm{\Sigma }}_{i=1}^p{B}_i{y}_{t-i}+{\epsilon }_t$ (1).

	$R_t$	${MV}_t$	${DT}_t$
$R_{t-1}$	0.0548	−0.5560	−4.2152***
	(1.2240)	(-0.7169)	(-4.4729)
$R_{t-2}$	−0.0446	−0.5553	−1.2796
	(-0.9614)	(-0.6916)	(-1.3116)
${MV}_{t-1}$	0.0053	−0.3198***	0.2481***
	(1.4921)	(-5.2396)	(3.3459)
${MV}_{t-2}$	−0.0004	−0.1746***	0.0893
	(-0.1269)	(-2.8612)	(1.2047)
${DT}_{t-1}$	−0.0020	−0.0414	−0.4924***
	(-0.6878)	(-0.8337)	(-8.1707)
${DT}_{t-2}$	−0.0006	−0.0550	−0.2482***
	(-0.2202)	(-1.1579)	(-4.2978)
Constant	−0.0086	−0.0846	0.6504
	(-0.2198)	(-0.1249)	(0.7894)
F statistic (p-value)
$H_o:{\mathrm{\Sigma }}_{i=1}^2{b}_{Ri}=0$	1.1429	0.5288	11.2691***
	(0.3197)	(0.5897)	(0.0000)
$H_o:{\mathrm{\Sigma }}_{i=1}^2{b}_{MVi}=0$	1.3652	14.2777***	5.5975***
	(0.2563)	(0.0000)	(0.0039)
$H_o:{\mathrm{\Sigma }}_{i=1}^2{b}_{DTi}=0$	0.2372	0.7669	34.3537***
	(0.7889)	(0.4650)	(0.0000)

Note: Same Note shown in Table 2.

Table 7.

The result of the VAR model for Period 2 (Feb. 2016–Dec. 2017). ${\mathrm{y}}_t={\left[R_t,{MV}_t,{DT}_t\right]}^{\prime }=A_0+{\mathrm{\Sigma }}_{i=1}^p{B}_i{y}_{t-i}+{\epsilon }_t$ (1).

	$R_t$	${MV}_t$	${DT}_t$
$R_{t-1}$	0.0032	2.5751**	1.4145
	(0.0697)	(2.1607)	(1.2215)
$R_{t-2}$	0.0229	1.6560	0.3299
	(0.4957)	(1.4018)	(0.2874)
${MV}_{t-1}$	−0.0002	−0.3645***	0.0714
	(-0.0788)	(-5.2654)	(1.0611)
${MV}_{t-2}$	−0.0009	0.0363	0.2175***
	(-0.3369)	(0.5252)	(3.2380)
${DT}_{t-1}$	0.0025	−0.0875	−0.5463***
	(0.9222)	(-1.2598)	(-8.0982)
${DT}_{t-2}$	0.0013	−0.2556***	−0.4655***
	(0.4829)	(-3.6800)	(-6.8978)
Constant	0.0547*	−0.0543	0.3865
	(1.7634)	(-0.0683)	(0.5003)
F statistic (p-value)
$H_o:{\mathrm{\Sigma }}_{i=1}^2{b}_{Ri}=0$	0.1255	3.3365**	0.7897
	(0.8821)	(0.0364)	(0.4546)
$H_o:{\mathrm{\Sigma }}_{i=1}^2{b}_{MVi}=0$	0.0580	17.4410***	5.2542***
	(0.9437)	(0.0000)	(0.0055)
$H_o:{\mathrm{\Sigma }}_{i=1}^2{b}_{DTi}=0$	0.4361	6.7799***	41.4397***
	(0.6468)	(0.0013)	(0.0000)

Note: Same Note shown in Table 2.

Next, stock return at time t-1 significantly negatively affects day trading volume at time t in Period 1, but the above phenomena might not exist in Period 2. Similarly, based on the statistics (F statistics) shown in Table 6, Table 7, we reveal that stock returns at time t-1 negatively affect day trading volume at time t in Period 1 instead of Period 2 but day trading volume at time t-1 would not impact stock returns at time t for both Periods. The result indicates that day trading volume at time t might be enhanced after inferior stock market performance at time t-1 in Period 1, implying that lower share prices might appeal to day traders, thus enhancing day trading volume. Furthermore, the market trading volume at time t-1 would not impact stock return at time t in Table 6, Table 7 but stock returns at time t-1 would impact market trading volume at time t in Table 7 instead of Table 6, indicating that better stock returns would enhance subsequent market trading volume after the relaxation. Similar results are revealed in the Granger F tests.

Furthermore, we reveal that the transmission between stock return and trading activities including market trading volume and day trading volume seems to change after the relaxation. As a result, we further explore whether the role of trading volume would change after the relaxation.

5.2. The dynamic effects of trading volume in subperiods

Similar to the issues examined in Section 4, we further examine whether stock return would be affected by trading volume through the channel of conditional volatility in these two subperiods. We reveal that while comparing the results shown in Table 3 (overall period) shown above, Table 8 shows that stock return would be (not) impacted by day trading volume through the channel of conditional volatility in Period 2 (Period 1). The result indicates that Hypothesis 2 is mainly supported by employing the data of Period 2 instead of Period 1, indicating that the relaxation of day trading may dominate this finding.

$${\mathrm{R}}_t={\beta }_0+{{\mathrm{\Sigma }}_{i=1}^2\beta }_{Ri}{R}_{t-i}+{\beta }_h\log h_t+u_t,u_t\sim N\left(0,h_t\right)$$

$$\log h_t=c_0+a\left|u_{t-1}\right|/\sqrt{h_{t-1}}+b\log h_{t-1}+d{u}_{t-1}/\sqrt{h_{t-1}}+\omega {Volume}_t+{\nu }_t$$ (2a)

Table 8.

The result of the EGARCH(1,1)-M model for subsamples.

	Volume: Market trading volume		Volume: Day trading volume
	Period 1	Period 2	Period 1	Period 2
${\beta }_0$	−0.0358	0.1185*	0.0094	0.4135***
${\beta }_{R1}$	−0.0360	0.0266	0.0315	0.0772
${\beta }_{R2}$	−0.0666*	−0.0354	−0.0376	0.0561
${\beta }_h$	−0.0242	0.1163*	0.0362	0.3618***
$c_0$	−0.0272	−0.4001***	−0.0519**	−0.3935***
$a$	0.0190	0.2097***	0.0487*	0.1976***
$b$	0.9765***	0.7843***	0.9768***	0.7569***
$d$	−0.1349***	−0.1957***	−0.0983***	−0.1885***
$\omega$	0.0355	0.0287***	0.0104***	0.0159***
Log-likelihood function	−556.4996	−415.0759	−600.3160	−438.8759
Model diagnosis
Ljung-Box Q(10)	6.2560	9.3240	5.6067	7.6105
McLeod-Li(10)	6.6624	8.9193	6.3981	10.0905

Note: Same Note shown in Table 3. Two sub-periods include period 1 before the relaxation (i.e, Jan. 2014–Jan. 2016) and period 2 after relaxation (Feb. 2016–Dec. 2017).

We further explore whether stock returns are impacted by expected (unexpected) trading volume through conditional volatility. Table 9 shows that stock return is significantly affected by unexpected trading volume instead of expected trading volume through the channel of conditional volatility, supporting Hypothesis 3.

$${\mathrm{R}}_t={\beta }_0+{{\mathrm{\Sigma }}_{i=1}^2\beta }_{Ri}{R}_{t-i}+{\beta }_h\log h_t+u_t,u_t\sim N\left(0,h_t\right)$$

$$\log h_t=c_0+a\left|u_{t-1}\right|/\sqrt{h_{t-1}}+b\log h_{t-1}+d{u}_{t-1}/\sqrt{h_{t-1}}+{\omega }_e{Volume}_t^e+{\omega }_{u0}{Volume}_t^{un}+{\omega }_{u1}{Volume}_{t-1}^{un}+{\omega }_{n}D\times {Volume}_t^{un}+{\nu }_t$$ (3a)

Table 9.

The result of the EGARCH(1,1)-M model with expected and unexpected volume for subsamples.

	Volume: Market trading volume		Volume: Day trading volume
	Period 1	Period 2	Period 1	Period 2
${\beta }_0$	−0.0219	0.2353***	0.0171	0.3563***
${\beta }_{R1}$	−0.0215	0.0292	0.0271	0.0513
${\beta }_{R2}$	−0.0849**	−0.0061	−0.0410	0.0391
${\beta }_h$	−0.0123	0.1606**	0.0545	0.2945***
$c_0$	−0.1022**	−0.4891***	−0.0463	−0.3394***
$a$	0.0675*	0.1804***	0.0396	0.1366***
$b$	0.9728***	0.7904***	0.9762***	0.8348***
$d$	−0.1248**	−0.1989***	−0.0957***	−0.1383***
${\omega }_e$	0.0165	0.0182	0.0061	−0.0015
${\omega }_{u0}$	0.0437***	0.0327***	0.0145***	0.0202***
${\omega }_{u1}$	−0.0047	−0.0033	0.0025	0.0036
${\omega }_n$	0.0041	0.0131**	0.0003	0.0077***
Log-likelihood function	−547.9371	−409.8850	−596.0431	−434.7968
Model diagnosis
Ljung-Box Q(10)	8.2031	7.5230	5.9229	7.7407
McLeod-Li(10)	13.5879	9.8453	7.3447	8.2188

Note: Same Note is shown in Table 4.

Furthermore, we incorporate the VIX into the conditional volatility of our model. We reveal that unlike the results shown in Table 5, Table 10 reveals that stock volatility would be affected by the trading volume with VIX at a rather higher level. In addition, Table 10 reveals that stock returns are more significantly impacted by day trading volume through the channel of conditional volatility in period 2. Further, Table 10 also shows that stock market volatility would be more enhanced by day trading instead of market trading volume in Period 2 when the VIX index falls into a rather high level as the coefficient of ${Volume}_t\times D_{VIXH,t}$ is more significant for day trading volume compared with market trading volume.

$${\mathrm{R}}_t={\beta }_0+{{\mathrm{\Sigma }}_{i=1}^2\beta }_{Ri}{R}_{t-i}+{\beta }_h\log h_t+u_t,u_t\sim N\left(0,h_t\right)$$

$$\log h_t=c_0+a\left|u_{t-1}\right|/\sqrt{h_{t-1}}+b\log h_{t-1}+d{u}_{t-1}/\sqrt{h_{t-1}}+\omega {Volume}_t+{\omega }_H{Volume}_t\times D_{VIXH,t}+{\omega }_L{Volume}_t\times D_{VIXL,t}+{\nu }_t$$ (4a)

Table 10.

The result of the EGARCH(1,1)-M model incorporating VIX for subsamples.

	Market trading volume		Day trading volume
	Period 1	Period 2	Period 1	Period 2
${\beta }_0$	−0.0162	0.2053**	0.0152	0.5314***
${\beta }_{R1}$	−0.0106	0.0433	0.0333	0.1069**
${\beta }_{R2}$	−0.0660*	−0.0365	−0.0366	0.0535
${\mathit{\beta }}_{\mathit{h}}$	0.0031	0.1304*	0.0452	0.4669***
$c_0$	−0.0281	−0.5225***	−0.0517**	−0.4194***
$a$	0.0177	0.2140**	0.0484**	0.1601***
$b$	0.9809***	0.6859***	0.9774***	0.7116***
$d$	−0.1333***	−0.1939***	−0.0987***	−0.1759
$\omega$	0.0334***	0.0238***	0.0102	0.0127***
${\mathit{\omega }}_{\mathit{H}}$	0.0051	0.0243**	0.0004***	0.0173***
${\omega }_L$	0.0124	−0.0027	0.0016	0.0013
Log-likelihood function	−535.7632	−411.9000	−600.2726	−435.1038
Model diagnosis
Ljung-Box Q(10)	5.1086	8.1052	5.6552	9.6459
McLeod-Li(10)	10.8880	7.1637	6.6785	7.5733

Notes: Same Note is shown in Table 5.

6. Concluding remarks

Day trading volume has considerably increased in TWSE after the relaxation (i.e., day trading stocks are permitted comprehensively at the beginning of 2016 in Taiwan). The above phenomenon indeed arouses our interest in further investigation. The objective of this research is to investigate the impact and profitability of day trading following the relaxation of day trading restrictions in Taiwan, using time series models such as VAR models and EGRACH models. As a result of offering many essential hypotheses about this relaxation, we arrive at the following crucial conclusions.

6.1. Conclusion and discussions

First, we demonstrate that a high market trading volume would result in a high day trading volume, which could be due to liquidity markets preferred by day traders, but a high day trading volume would not result in a high market trading volume, which could be due to speculative markets not preferred by other market participants. The foregoing findings suggest that the indications of increased day trading volume may be mixed, with liquidity risk lowered [57,81] and speculative markets suspected [86,87].

Second, contrary to our expectations, the VIX index rises at a high level, yet day trading profits increase following the relaxation. A higher VIX index may indicate a volatile market, which may increase the intraday spread of a stock that is good for day trading. In other words, while investors' sentiments are aroused by fear due to market panic [[88], [89], [90]], day traders may generate profits during market panic [91] by widening the spread of stock [92] or increasing the probability of buying low and selling high [93] often occurs during this period, thereby likely enhancing the profitability of day traders [94].

Third, we reveal that, relative to unexpected market trading volume, unexpected day trading volume would have a greater effect on stock returns through the channel of stock volatility. We conclude that unexpected day trading volume may be more unexpected than unexpected market trading volume, hence heightening stock market volatility and boosting stock returns. The aforementioned findings are consistent with the likelihood that stock markets are influenced by unexpected information regarding returns, volume, and volatility [57,95,96].

6.2. Policy and managerial implications

Moreover, this study has two valuable implications. First, by disclosing that the profitability of day trading would be enhanced after the relaxation, we argue that quite a few day traders might have insider information, abundant capital, and even expert trading skills, which might be somewhat different from day traders likely regarded as losers in the relevant studies.

Second, the authorities might evaluate the relaxation by taking our revealed results into account, as revealed that stock markets seem enhanced after the relaxation. However, we argue that some well-informed investors may be able to profit from market asymmetry caused by regulatory changes (i.e., relaxing day trading restrictions). Based on information asymmetry being unbeneficial for investors, especially for individual investors, we thus argue that while authorities implement the new regulations, authorities should make efforts on enhancing capital markets and reducing information asymmetry since sound capital markets would be the sustainable goal of capital markets.

6.3. Limitations and future research

Besides, we collect daily data for day trading volume from the web of the Taiwan Stock Exchange due to the data only released on this web; however, we might not able to collect complete daily day trading volume data after 2017 due to the restriction of the website, which may be the limitation of this study. Even so, we endeavored to collect data after 2017. Despite the fact that some data are missing, we find that our revealed results using the data (i.e., 2016/02/01–2017/12/31) may not appear for using the data after 2017. We infer that when new regulations are implemented or new capital markets are established, those who are able to grasp the information ahead of time and even exploit information asymmetry may profit. However, such phenomena may not last long because capital markets may approach market efficiency as information asymmetry decreases as more people become aware of the new regulation. Similar results have been discovered in China stock markets [97] and China stock index futures markets [98]. As a result, if shifting laws or regulations in capital markets occur in the future, we might consider their future studies and then investigate whether exploiting profitable opportunities would be temporary only or whether those who are sensitive to changes in capital markets would profit instead of other investors.

Author contribution statement

Wan-Hsiu Cheng, Ph.D.: Conceived and designed the experiments; Performed the experiments; Analyzed and interpreted the data; Contributed reagents, materials, analysis tools or data; Wrote the paper.

Yuhsin Chen, Ph.D; Yensen Ni, Ph.D: Conceived and designed the experiments; Performed the experiments; Analyzed and interpreted the data; Wrote the paper.

Paoyu Huang, Ph.D.: Performed the experiments; Analyzed and interpreted the data; Contributed reagents, materials, analysis tools or data; Wrote the paper.

Mei-Chu Liang, Ph.D. Candidate: Analyzed and interpreted the data; Contributed reagents, materials, analysis tools or data; Wrote the paper.

Funding statement

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Data availability statement

Data will be made available on request.

Declaration of interest’s statement

The authors declare no competing interests.