In this section we present our results and analyse them.
Results
See Tables 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 and 11.
Table 1 Returns for all pairs trades (rALL), pairs trades during an operational event (rPTE), and pairs trade event returns (rE) Table 2 All pairs trades (rALL) distribution measures, risk measures and test statistics Table 3 Change in values and test statistics: pairs trades during an operational event rPTE compared to all trades (rALL) Table 4 Change in value at risk and test statistics: pairs trades during an operational event rPTE compared to all trades (rALL) Table 5 Change in values, risk measures and test statistics: pairs trade event returns rE compared to all trades (rALL) Table 6 Pairs trade return during an operational event (rPTE) by operational event type Table 7 Change in pairs trades during an operational event (rPTE) compared to all pairs trades (rALL) values: by operational event type Table 8 Pairs trade returns during an operational event (rPTE): by business line Table 9 Change in pairs trade returns during an operational event (rPTE) compared to all pairs trades (rALL): by business line Table 10 Categories for operational event types and abbreviations Table 11 Categories for business lines and abbreviations Analysis
In Tables 1, 2, 3, 4, 5, 6, 7, 8 and 9 we present the results for all our PT in our study; Tables 10 and 11 detail our OE types and business lines used in our study. In Table 1 the ‘All’ results denote the results relating to all 11,648 PT, including all PT during and outside any OE; the ‘PTE’ results denote PT returns during OE only, and ‘Event’ results denote rE returns (that is PT return due to the event itself). The figures in all the tables refer to values over the 1 week duration for the PT; however the annualised values are the weekly values converted to their annual equivalent e.g. weekly expected returns are converted to their annual equivalent.
We note in passing that in order to calculate rE requires rALL (given in tables) and rNPT; mean rNPT was calculated to be − 0.05% per trade. On an annualised basis this is approximately equivalent to a magnitude of 2.6%, which is also approximately equal to the risk free rate during the period of study (the US interest rates fluctuated between 1 and 6%, with an approximate average of 3%). Hence the rNPT return is negligible and not economically significant. We also note that we would not expect rNPT to be economically significant because rNPT is defined as the return during no event, and as discussed previously such returns would be expected to be negligible.
The OR was calculated using risk measures: VaR (value at risk) at different quantiles, SD (standard deviation), the third and fourth moments (skewness and kurtosis, respectively). The OR associated with the results in Table 1 are presented in Tables 2, 3, 4 and 5; Table 2 provides the risk measures under rALL results, in Tables 3, 4 and 5 we present the change in the risk measure’s value under the rPTE and rE when compared rALL risk measure results.
To test the empirical distributions for following a Normal distribution but with the same SD, we conducted Kolmogorov–Smirnoff tests. The Normal distribution test was conducted to determine the empirical distribution’s likelihood of outliers in its distribution (compared to a Normal distribution) but also a non-Normal distribution typically increases the likelihood of incorrectly modelling and estimating risk. In Table 2 a Kolmogorov–Smirnoff test is executed, to test rALL distribution against a Normal distribution with 0 mean but with same SD (3.65); we denote the Normal distribution by N(0, 3.65). Similarly, in Table 4 a Kolmogorov–Smirnoff test is executed to test rPTE distribution against N(0, 3.27), a Normal distribution with 0 mean but with same SD as rPTE distribution (3.27).
In Table 3 we conduct a two sample equal variance F test, to test the similarity in variance between the rPTE and rALL distributions; this was to determine the similarity in risk between rPTE and rALL on a SD risk measurement basis. In Table 3 we also provide the rALL expected return sampling confidence intervals, so that the range of potential rALL expected return values (for the given sample size) are known (at the given confidence intervals). This would enable us to determine if the expected return for rPTE is statistically different to rALL, or if rPTE is different.
In Table 5 we provide the change in values for risk measures and other measures in comparing rALL against rE results. We also conduct a Kolmogorov–Smirnoff test for the rE distribution against N(0, 3.27), a Normal distribution with 0 mean but with same SD as rE. The PTE results of Table 1 (or Tables 3, 4) are analysed further by OE type and business line in Tables 6 and 8, respectively; the PTE results of Table 2 are also analysed further by OE type and business line in Tables 7 and 9, respectively.
In Table 1, we observe there is significant variation in expected PT returns by categories; expected PT returns are 2.28%/year for All trades (PT for rALL), 6.19%/year for PTE trades (PT trades for rPTE) and 8.79%/year for event trades (rE). The difference in average returns in rALL, rPTE and rE is also supported by the annualised median returns in Table 1, which is considered a more robust indicator of average values than the mean. The interquartile range for annualised returns for rALL is approximately 15% more than the range for rPTE or rE. As the interquartile range is a standard measure of variability or dispersion, therefore the probability distribution for rALL is more dispersed than for rPTE or rE. Also, recall that rE represents the PT return due to the OE itself, hence the OE itself leads to a PT return that is almost four times the rALL returns.
Table 1 results show that PT returns increase by more than double during an OE compared to no OE occurring during the PT, that is rPTE is more than double rALL. For statistical robustness, we calculated rALL expected return’s sampling confidence intervals, at the 95% an 90% quantile values, and the values are given in Table 3. Therefore the 0.08% increase in expected return from rALL to rPTE is outside the range of any sampling error, hence it is a statistically significant increase in expected returns.
Table 1 results clearly demonstrate that OE significantly impact relative firm value. In fact the rE annualised expected returns are almost equivalent to the average annual returns on the stockmarket (approximately 10%/year). The impact of OE upon firm value, and more importantly relative firm value, implies that financial technology has an important role in competitive advantage and therefore strategy. For example financial technology could provide insight into competitive advantage and relative firm value being lost from poor process management and execution, or poor workplace practises (e.g. unnecessary losses due to overriding trading limits). The results also suggest that any operational gains through FPAS could lead to significant increases in relative firm value.
The contribution of financial technology to relative firm value also occurs through them having a key role in enabling any strategic goal. For example, firms frequently focus on good customer service to gain strategic advantages over competitors [76]. The financial technology would enable strong customer service by ensuring there is low external fraud, internal fraud and strong business practices. In an industry sector such as banking, the operational advantage of a competitor can be a significant and unique selling point.
In Table 2 we conduct the Kolmogorov–Smirnoff test to assess the similarity of the rALL distribution to N(0, 3.65). As can be seen in Table 2 the test statistic is Dcrit = 0.07 and does not exceed the critical values at the 5% and 10% significance levels (0.013 an 0.011, respectively). Consequently, we cannot reject the hypothesis that the rALL distribution is different to a Normal distribution [N(0, 3.65)]. This implies that the rALL distribution is not an unusually shaped distribution, that is neither heavily bias towards positive or negative values (as might be the case with other distributions), hence easier to model and risk manage.
Table 2 provides some useful results. Firstly, OR measured on the basis of VaR is not negligible; at the VaR 99% level we have − 10.22% and − 3.99% at the 90% level, both relating to a 1 week time interval. As the approximate annual return on the stock market is 10%/year, the VaR results at the 99% level imply that an average share price growth can be erased in a period of 1 week. Although the 99% level is a highly unlikely event, the results imply that OR can cause significant losses to relative firm value. Hence it is important for operational systems not to be complacent about a lack of significant losses to relative firm value, for when they occur they can be large in magnitude.
In Table 3 we conducted a two sample equal variance F test upon rPTE and rALL distributions’ variances, to determine if both distributions have the same risk, statistically, on a SD risk measurement basis. The test statistic is Fcrit = 1.24 and this exceeds the significance values at the 5% and 10% significance levels (1.15 and 1.12, respectively). Therefore we can reject the hypothesis that the rPTE and rALL distributions have equal variance. Hence the OR and relative firm value changes during OE will be different to those during All trades, implying that financial technology impact the OR and relative firm value in companies.
Table 3 results show that the skewness difference between the rPTE and rALL distributions is negligible (close to 0), hence both distributions are fairly symmetric and do not suffer from skewness risk (underestimation of risk). However from Table 3 we observe that there is significant difference in kurtosis between rPTE and rALL distributions (− 2.44), rALL has almost double the kurtosis. A lower kurtosis implies that the rPTE distribution tends to concentrate around mean more than the rALL distribution and so the rPTE distribution values can be considered more predictable. This is useful for operational analysis because it means losses associated with financial technology would be more predictable, due to the lower kurtosis.
Table 4 results show the difference in VaR between rALL and rPTE and there is a change of 0.14–1.70% in VaR value, which is approximately a 10–20% difference; the change is even greater between rALL and rE, which is a change of 0.19–1.76% in VaR value. The lower OR in rPTE and rE compared to rALL on a VaR risk measure basis is also supported by the SD risk measure. This is an unexpected result, since returns increase during an OE yet risk decreases (typically we expect return to increase with risk under the standard assumption of risk aversion [77]). Therefore not only do operational factors have a contribution to relative firm value, but they also do so at lower risk. Such information is useful for operational products because it suggests that such products can contribute to relative firm value without having to incur substantially more risk.
In Tables 4 and 5, the Kolmogorov–Smirnoff tests were undertaken to test rPTE and rE distributions’ similarities against a Normal distribution with 0 mean but with the same respective SD. As both distributions have identical SD, both distributions were tested against N(0, 3.27). For both tables the test statistic is given as Dcrit = 0.08 and both distributions reject the hypothesis that either are statistically similar to N(0, 3.27) at the 10% significance level, as 0.08 exceeds D10% = 0.07. However, at the 5% significance level it less certain if both distributions differ from N(0, 3.27) because Dcrit = D5% = 0.08. Therefore, whereas the rALL distribution tends to follow a Normal distribution, the PT distributions during OE can be considered statistically less Normal. This is useful to know because non-Normal shaped distributions tend to be a common cause of mis-estimation of losses, as they are harder to model and risk manage.
Table 5 results relate to the rE distribution and we notice they provide similar values for the same measures in Tables 3 and 4. This is partly to be expected because rE is related to rPTE by the equation rPTE = rNPT + rE, hence risk and return values will be related. However, the high amount of similarity also implies that the majority of rPTE values can be attributed to rE. This is useful for financial technology design because it implies relative firm value is driven more by specific OE, hence design should focus around such events rather than generic operational issues.
Tables 6 and 7 decompose the OE returns and risk measurements by OE type, as listed in Table 10. In Table 6 we have the PT returns by OE type and we observe significant variation in returns: OPR has the highest annualised expected return (208.90%), the lowest annualised expected return is WPS (− 143.52%) and the average annualised PT return is 6.24%. The annualised median returns also support that OPR has the highest PT return (although at a slightly lower value of 208.90%), however BP has the lowest return at − 43.55% (instead of WPS). We also notice that there are varying differences between median and expected return values e.g. PME and WPS have substantial differences whereas OPR and BP are similar.
Table 6 provides important insights. Firstly, as mentioned previously, the results support the analysis that relative firm value (or equivalently competitive advantage) is strongly dependent on the type of OE, with wide variation in relative firm values between OE. The wide variation in not just restricted to the expected return but also the interquartile range varies significantly by OE. Secondly, the magnitude of the returns in Table 6 is significant: given that average stock market returns are approximately 10%/year, the OPR return of 226.20% and the WPS return of − 143.52% the OE represent a high and wide range of returns. Hence financial technology can maximise competitive advantage by strategically reallocating resources to focus on specific OE types, rather than focussing on less critical operations or having a generic operations focus. Moreover the high magnitude supports the point that financial technology have a key role in business strategy.
In Table 7 we provide OR measurement by event type with a range of risk measures, specifically SD, VaR at different quantiles, 3rd and 4th moments. Similar to Table 6, we notice that there is significant dependence and variation in OR by event type, at all different measures of OR. Intuitively, we would expect OR to be considerably dependent on the OE type, as some operations tend to be more risky than others e.g. external fraud (event FEX) is considered more risky than process management errors (event PME).
Table 7 results give important observations into OR. Firstly, the wide variation in OR implies that financial technologies should carefully concentrate on operational areas to minimise risk, rather than having a generic OR management strategy. A strategic reallocation may lead to risk reduction e.g. shifting focus from FEX to FIN (internal faud). Secondly, given that optimal contributions of relative firm value and OR differ in terms of OE, one needs to examine both aspects to achieve the correct contribution to firms. There is wide variation in magnitude with OE in terms of return and OR (the change in OR under a VaR 99% risk measure is 0.04% for FEX but 10.07% for OPR).
In Table 7 we notice that there is significant variation in skewness and kurtosis; OPR, PME and BP are more positively skewed (compared to the rALL distribution), whereas FIN, FEX, WPS and PDA are more negatively skewed. Hence the positively skewed distributions will lead to more negative returns and similarly the negatively skewed distributions will give more positive returns. In terms of kurtosis, Table 7 implies that the OE tend to have a lower kurtosis compared to the rALL distribution, hence the OE distributions tend to be less peaked and so are more spread out. The variation in kurtosis and skewness leads to kurtosis and skewness risk, namely that skewness and kurtosis can lead to over or under estimation of risk. This is important as one may be more exposed to incorrectly estimating risk depending on the type of OE they are engaged in.
In Tables 8 and 9 the PT results during OE (rPTE) are categorised by business lines. In Table 8 we notice that the magnitude of the expected PT returns is significant for all business lines. This is also supported by the median returns, which are considered a more robust indicator of average values, in that they are also a similar magnitude to the expected returns. Intuitively, we expect MIS systems to have a significant contribution to relative firm value across all business lines because MIS nowadays play an important function in all aspects of all business lines in banks. Similarly, the results imply that FPAS would be beneficial across all business lines and this may account for the fact that such products being available across a range of banking business lines nowadays.
In Table 8 we notice that the variation in interquartile ranges, medians and expected returns are substantially lower than in Table 6. The interquartile range in Table 8 is an average of 158% whereas in Table 6 it is 211%; the variation in expected returns in Table 8, is approximately half the variation in Table 6. Therefore we can deduce that the variation in relative firm value is more due to the OE type rather than the business line origin. This is an important observation because it implies strategically allocating financial technologies along business lines is not as effective as allocating along particular OE types.
In Table 9 the OR measurement is given in terms of SD and VaR at various quantiles and the OR values are similar in magnitude to the OR values in Table 7. Hence we can infer that OR is not substantially influenced by business line more or less than the OE type in terms of magnitude. However, within the set of different business lines in Table 9 we notice that there is substantial variation in risk; at a VaR 99% level the change in VaR value (compared to the rALL distribution) is almost 5 times greater in CRT than in CMB. Therefore in strategically resourcing we cannot purely focus on OE but also must take into account the business line to optimise contribution to relative firm value and competitive advantage.
In Table 9 the variation in skewness and kurtosis between business lines is higher than the variation in both for OEs in Table 7; in Table 9 the average values for skewness and kurtosis are − 0.07 and − 3.8 (respectively) whereas in Table 7 they are 0.01 and − 3.4 (respectively). Therefore in terms of skewness and kurtosis risk, the business lines present a greater challenge in correctly estimating risk compared to OE. This is important as the results imply that firms are more exposed to incorrectly estimating risk depending on the business line, rather than the OE. Hence to minimise risk, firms may wish to change the business line for financial technology to reduce risk.