誰的算法這么說：公司類型、信任和專業知識的看法以及金融機器人咨詢的接受度之間的關系.pdf

《誰的算法這么說：公司類型、信任和專業知識的看法以及金融機器人咨詢的接受度之間的關系.pdf》由會員分享，可在線閱讀，更多相關《誰的算法這么說：公司類型、信任和專業知識的看法以及金融機器人咨詢的接受度之間的關系.pdf（34頁珍藏版）》請在三個皮匠報告上搜索。

1、Whose Algorithm Says So:The Relationships Between Type of Firm,Perceptions of Trust and Expertise,and the Acceptance of Financial Robo-AdviceCarlos J.S.Loureno1,3Benedict G.C.Dellaert2,3Bas Donkers2,31ISEG&SOCIUS/CSG,University of Lisbon2Erasmus University Rotterdam3NETSPARThis presentation was supp

2、orted by FCT,I.P.,the Portuguese national funding agency for science,research and technology,under the Project UIDB/04521/2020.The study was funded by NETSPARs grant to the Interactive pension communication and decision-making project.9th International Pension Research Association Conference12 June

3、2024,OECD,Parisrobo-advice and robo-advisors:a product-matching marketing perspective since early 2010s:widespread automated interactive financial advice,namely on retirement and pension planning by 2025 robots expected to manage$16 T(14.7 MM)assets(Deloitte 2016)average time on smartphones(in US)ap

4、prox.5 hours/day and increasing attractive for the industry digital/automated marketing allows lowering costs and thus coming up with a new complementary and cheaper sales channelCora private by insurer and pension provider Liverpool Victoria(UK)robot marketed as“much faster than a human advisor”pro

5、fit orientedInvestment Balance private by financial services provider Centraal Beheer(NL)similar to Achmeas and SBZsBlue Zone public or state-sponsored developed with the University of Minnesota School of Public Health(healthy-)life expectancy algorithm customized recommendations non-profit oriented

6、ESPlanner(now MaxiFi)public or state-sponsored developed by Boston University economics Prof.Laurence Kotlikoff“robo-optimizing”lifetime financial planning tool since 1999 non-profit oriented(initially)in theory,it shouldnt matter upon receiving the same input information personalized recommendation

7、s generated by automated tools of different organizations should be the same and thus,should be equally accepted by consumershowever egocentric discounting irrationally overweight own opinion relative to that of an unbiased advisor(Harvey and Fischer 1997;Yaniv and Kleinberger 2000;for a review see

8、Bonaccio and Dalal 2006)algorithm aversion irrationally discount unbiased advice generated by computer algorithms(e.g.,Dietvorst et al.,2015,Goodwin et al.,2013)principle-agent problem on the background incentives of advisor(“agent”)may not align with those of advisee(“principal”)underscoring firms

9、ability to elicit proper individual risk preferences(Donkers,Loureno,and Dellaert 2012)which raises at least two main(research)questions does the type of firm providing robo-advice affect advice acceptance?if so,which advisor firms are best suited to provide automated pension advice,i.e.,whose advic

10、e is most accepted?what are the underlying drivers of the different acceptance rates between these firms?and how do they play a role?in our study we propose to use firm characteristics that signal consumers different incentives to provide advice and how(un)aligned they may be with those of consumers

11、 to focus on two such firm characteristics and thus study four types of firms:for-vs.not-for-profit orientation product provider vs.advisor-only role in the sales channelin our study we propose only the four types of firms were made explicit manipulation check was conducted on a separate online stud

12、y(N=201)profit orientation:insurers&commercial comparison websites pension funds&information websites of the government product providers:insurers&pension funds commercial comparison websites&information websites of the governmentin our study we propose to look at how the different types of advisor

13、firms are perceived to be trustworthy(e.g.,Sniezek and Van Swol 2001;cf.Prahl and Van Swol 2017)experts(e.g.,Sniezek,Schrah,and Dalal 2004;cf.Prahl and Van Swol 2017)because looking after and following an advice implies a shared responsibility for the outcomes(Harvey and Fischer 1997)professional ad

14、vice isnt considered manipulative or invasive but a means to improve participants decisions(Schrah,Dalal,and Sniezek 2006;Yaniv 2004)in our study we propose a sequentially mediated process by which a firms profit orientation&role in the sales channel through their effect on consumer perceptions of a

15、 firms expertise and trustworthiness which,in turn,affect the consumers satisfaction using the automated algorithm/robot determine the acceptance of the robo-advice3 challenges:3 challenges:how to control for and compare to the“no-advice”case?we design four(explicit)advice treatments and an(implicit

16、)no advice treatment consumer gives herself the advice that the firm would have communicated firm is only facilitating the use of the algorithm on which the consumer herself generates the advice allows testing baseline effect of automated firm-advice“the firm has created a new retirement simulator,a

17、nd you will have to indicate when you want to retire and how much risk you want to take with your pension investmentsand then the firm will give you appropriate advice about your pension investments based on your preferences”“the firm has created a new retirement simulator that you can use to help y

18、ourself make your choice and choose when you want to retire and how much risk you want to take with your pension investments and then try out various options and decide for yourself which one suits you best”3 challenges:3 challenges:how to test web of relationships?we use an econometric structural e

19、quation model(SEM)that estimates conceptualized relationships simultaneously(Iacobucci,2008,Zhao et al.,2010)a SEM model handles estimation uncertainty jointly and efficiently3 challenges:3 challenges:what automated algorithm to use to generate unbiased individual advice?we developed the pension bui

20、lder robot and algorithm based on sound economics and previous research(Goldstein and Sharpe 2000;Goldstein et al.2008)users learn&experience risk-return tradeoffs interactively on a graphical online interface risk represented as frequencies(2 in 100)rather than percentages(2%),which improves unders

21、tanding of probabilities(Fagerlin,Zikmund-Fisher,and Ubel 2011)pretested in several rounds with employees at Netspar partner organizations and novices3 challenges:what automated algorithm to use to generate unbiased individual advice?user builds preferred income distribution with a slider based on t

22、he EU model and a CRRA,the algorithm uses constructed preferences to return a numerical estimate of an individuals attitude towards risk(the“lambda”;utility function curvature)and corresponding expected returns in three scenarios:optimistic,median,pessimistic(NL)choose the possible outcomes for your

23、 pension the number of times in 100 scenarios in which your income is as high as:your net monthly pension(in Euros)choose the possible outcomes for your pension the number of times in 100 scenarios in which your income is as high as:your net monthly pension(in Euros)choose the possible outcomes for

24、your pension the number of times in 100 scenarios in which your income is as high as:your net monthly pension(in Euros)measurements advice acceptance:“on a 0%100%probability scale,how likely are you to follow the online advice provided to you”(Elrod,Louviere,and Davey 1992)6-item scale for perceptio

25、ns of expertise&3-item scale for perceptions of trustworthiness(1=totally disagree;7=totally agree)interaction satisfaction with the robot(1=very dissatisfied;7=very satisfied)age,gender,income,educ.,user expertise(1=totally disagree;7=totally agree)data SSI collected data in NL from representative

26、sample of respondents(if belonging to working population and worked min 12h/week)N=1,649 respondents(6,473 started the study)38.1%females;17.5%HEduc after one item of the perceptions of expertise scale was dropped(it loaded also on the trust scale),the Cronbachs alphas were 0.97 for both scalesresul

27、ts:type of firm&satisfaction type of firm directly impacts satisfaction with the firms robot:pension providers less satisfactory than advisors-only(?=?0.183;p?.01)for-profits more satisfactory than not-for-profits(?=?0.270;p?.001)results:type of firm,expertise,trust a profit orientation is a double

28、jeopardy:negative impact on consumer perceptions of both expertise and trustworthiness for-profits considered less trustworthy(?=?0.491;p?.001)for-profits seen less as experts(?=?0.224;p?.001)to make things worse,expertise positively associated with trust(?=?0.843;p?.001)is carried over to satisfact

29、ion interacting with the robot:expertise increases satisfaction(?=?0.296;p?.001)trust increases satisfaction(?=?0.449;p?.001)a product provider is a double-edge sword:though pension providers are seen more like experts(?=?0.566;p?.001)they are trusted less than advisors-only(?=?0.378;p?.001)results:

30、type of firm,expertise,trust a profit orientation is a double jeopardy:negative impact on consumer perceptions of both expertise and trustworthiness for-profits considered less trustworthy(=|0.491|;p?.001)for-profits seen less as experts(=|0.224|;p?.001)to make things worse,expertise positively asso

31、ciated with trust(?=?0.843;p?.001)is carried over to satisfaction interacting with the robot:expertise increases satisfaction(?=?0.296;p?.001)trust increases satisfaction(?=?0.449;p?.001)a product provider is a double-edge sword:though pension providers are seen more like experts(=|0.566|;p?.001)the

32、y are trusted less than advisors-only(=|0.378|;p?.10&?=?0.148;p?.10,respectively)its fully mediated by(+)effect of both expertise&trust on acceptance(?=?3.272;p?.001&?=?1.185;p?1.890)its partially mediated by(+)effect of satisfaction on acceptance(?=?9.151;p?advice acceptance+9.2 pp all else constan

33、t pension advisors must ensure increasingly heterogeneous consumers and in particular older consumers closer to retirement are satisfied with automated(AI)tools online older consumers less satisfied with automated tool to generate pension advice(?=?0.005;p?.05)results&implications“least-trusted”for-

34、profits and product providers in particular may benefit from knowing that older consumers perceive firms as less trustworthy(?=?0.011;p?.001)challenge among female consumers who also perceive themselves as having lower user expertise(?=?0.546;p?.001)although more educated consumers more inclined to

35、accept robo-advice,they trust online pension advisors less(?=?3.008;p?.01);?=?0.170;p?.01)main implications robo-advice most likely to be accepted pension fund(high expertise&trust)insurance firm(high expertise,low trust)government-sponsored comparison website(high trust,low expertise)privately owne

36、d comparison website(low on expertise&trust)the 5.4 p.p.higher advice acceptance that pension funds enjoy(for same robot of a privately owned comparison website)may represent as much as$38.5 per consumer seeking advice(cost of typical session with human advisor is approx.$712 in UK;Delloite 2017)lim

37、itations and future research testing in a real world setting interactions may need to be more extensive(does the consumer has private savings or investments?)and may need updating from time to time consequences of ensuing endogeneity of robo-advice:automated(AI)algorithms will learn consumer preferences based on input of consumers,which itself depends on expected returns from the advice!.