Over the years I have spent lots of time on so­cial me­dia ar­gu­ing that the Video As­sis­tant Ref­er­ee Sys­tem isn’t fit for pur­pose. There is of course the usu­al push­back that the tech­nol­o­gy al­lows for the cor­rect de­ci­sion and every­one should be in favour of hav­ing the cor­rect de­ci­sions in a foot­ball match, es­pe­cial­ly when the stakes are high.

So­cial me­dia isn’t the place to re­al­ly ex­plain the is­sue and fur­ther my main el­e­ment of this dis­cus­sion comes from an un­der­stand­ing of how a stock mar­ket func­tions and then ap­ply­ing that in­sight to what takes place on a foot­ball field.

What every­one needs to grasp at the on­set is that you can make a de­ci­sion that abides by the rules and which is cor­rect but if that de­ci­sion is an ar­bi­trary one it los­es le­git­i­ma­cy. Think of the po­lice of­fi­cer who stops you for speed­ing while an­oth­er car is dri­ving faster over­tak­ing you.

At its core VAR is a hu­man de­signed, se­lec­tive, tech­no­log­i­cal high res­o­lu­tion re­view sys­tem. Now we will ex­am­ine whether it makes the sport more le­git­i­mate, more con­sis­tent and more ro­bust or con­verse­ly, whether it cre­ates a more so­phis­ti­cat­ed form of un­cer­tain­ty and sub­jec­tiv­i­ty while call­ing it ac­cu­ra­cy.

As a fi­nan­cial con­sul­tant, I spend much of my pro­fes­sion­al life help­ing peo­ple un­der­stand the dif­fer­ence be­tween a num­ber and a truth. Fi­nan­cial sys­tems pro­duce num­bers all day: prices, yields, spreads, re­turns, volatil­i­ty, cor­re­la­tions.

These can be use­ful but they are al­so dan­ger­ous pre­cise­ly be­cause they ap­pear sci­en­tif­ic. A mod­el can be pre­cise and still be wrong about the thing that mat­ters. That dis­tinc­tion is now cen­tral to our dis­cus­sion.

A use­ful start­ing point is the idea of a com­plex adap­tive sys­tem. A com­plex sys­tem is not mere­ly some­thing with many parts. A watch is com­pli­cat­ed, but it is not com­plex. Its parts are de­signed to be­have in pre­dictable ways. A com­plex adap­tive sys­tem is dif­fer­ent. It is made up of in­ter­act­ing agents whose be­hav­iour changes in re­sponse to one an­oth­er. The whole sys­tem pro­duces out­comes that can­not be un­der­stood by ex­am­in­ing the parts in iso­la­tion. In these sys­tems the agents adapt their rules and strate­gies as they in­ter­act.

The stock mar­ket is the ob­vi­ous ex­am­ple. A mar­ket is not a ma­chine that pas­sive­ly re­veals price. It is a liv­ing con­test among in­vestors, traders, al­go­rithms, reg­u­la­tors, cen­tral banks, an­a­lysts, jour­nal­ists, cor­po­rate ex­ec­u­tives and thou­sands of savers. Each par­tic­i­pant re­sponds to prices, and prices re­spond to par­tic­i­pants. The ob­ser­va­tion changes the thing ob­served.

A ris­ing share price may at­tract buy­ers, which push­es the price high­er. A falling mar­ket may pro­duce trig­gers that in­creas­es the sell­ing. Liq­uid­i­ty that ap­pears abun­dant in qui­et con­di­tions can van­ish un­der stress. This is why mod­ern fi­nan­cial analy­sis in­creas­ing­ly treats mar­kets as com­plex, in­ter­con­nect­ed sys­tems with feed­back loops as op­posed to equi­lib­ri­um ma­chines bal­anc­ing sup­ply and de­mand.

Foot­ball

Foot­ball is al­so a com­plex adap­tive sys­tem. In a match there is a dy­nam­ic ecol­o­gy of play­ers, coach­es, of­fi­cials, crowd pres­sure, tac­ti­cal in­cen­tives, fa­tigue, emo­tion, rep­u­ta­tion, time, score­line and in­sti­tu­tion­al pro­to­col. Every de­ci­sion changes be­hav­iour. A le­nient ref­er­ee in­vites more phys­i­cal­i­ty. An ear­ly yel­low card changes a de­fend­er’s risk ap­petite. A de­layed off­side flag changes how as­sis­tants, de­fend­ers and goal­keep­ers be­have. VAR it­self changes how play­ers ap­peal, how ref­er­ees de­fer, how coach­es com­plain and how fans ex­pe­ri­ence a goal.

Ap­ply­ing tech­nol­o­gy is an in­ter­ven­tion in­to a liv­ing sys­tem. The in­ter­ven­tion changes the sys­tem.

VAR was in­tro­duced to solve a gen­uine prob­lem. Foot­ball is low scor­ing, so a sin­gle in­cor­rect goal, penal­ty or red card can dom­i­nate a match. In risk lan­guage, these are tail events: rare rel­a­tive to the to­tal num­ber of ac­tions in a game, but dis­pro­por­tion­ate­ly con­se­quen­tial. It is ra­tio­nal to want a safe­guard against the cat­a­stroph­ic of­fi­ci­at­ing er­ror.

VAR as a tail risk con­trol sys­tem, claims to pre­vent the worst vis­i­ble mis­takes in the most con­se­quen­tial mo­ments. This sounds good on pa­per but it al­so cre­ates the cen­tral prob­lem. VAR does not scru­ti­nise foot­ball even­ly. In prac­tice, the mi­cro­scope comes out around goals, penal­ties and red cards, while the larg­er flow of the match re­mains gov­erned by or­di­nary hu­man per­cep­tion and judg­ment.

This cre­ates an asym­me­try. A foul that pre­vents a promis­ing at­tack may dis­ap­pear in­to the flow of the match. A small­er in­fringe­ment by the at­tack­ing team sec­onds be­fore a goal may be foren­si­cal­ly ex­am­ined and used to can­cel the goal. The suc­cess­ful at­tack is au­dit­ed. The il­le­gal­ly de­stroyed at­tack may not be. The sys­tem has cre­at­ed two stan­dards of re­al­i­ty: or­di­nary re­al­i­ty for most of the match, foren­sic pre­ci­sion for se­lect­ed de­ci­sive mo­ments.

Apre­ci­ate that pre­ci­sion is not the same as ac­cu­ra­cy, and ac­cu­ra­cy is not the same as jus­tice, which is a pre­requsite to le­git­i­ma­cy. In foot­ball terms, a sen­sor may an­swer one nar­row ques­tion: did the ball reg­is­ter a con­tact? It does not, by it­self, an­swer the larg­er ques­tions: was that con­tact legal­ly rel­e­vant, ma­te­ri­al­ly sig­nif­i­cant, con­sis­tent with the pur­pose of the law and pro­por­tion­ate­ly de­ci­sive?

A crud­er sys­tem miss­es things, so the al­lure is to­wards more sen­si­tive tech­nol­o­gy which al­so de­tects more noise. A hy­per sen­si­tive sys­tem finds things that may have no mean­ing­ful foot­balling con­se­quence. The process forces the de­vel­op­ment of hu­man de­signed fil­ters: how large must the sig­nal be, what counts as con­tact, how is vi­bra­tion dis­tin­guished from touch, how are ball da­ta and play­er track­ing da­ta syn­chro­nised? These are not pure­ly tech­no­log­i­cal ques­tions. They are sub­jec­tive hu­man judg­ment ques­tions moved in­to a tech­ni­cal lay­er.

Foot­ball now risks cre­at­ing a tail risk in­side its tail risk con­trol sys­tem. VAR was sup­posed to pro­tect the game from the rare cat­a­stroph­ic er­ror. Now you have the tail risk of a tour­na­ment-defin­ing goal can­celled by a mi­cro­scop­ic event that no play­er, ref­er­ee or spec­ta­tor ex­pe­ri­enced as mean­ing­ful foot­ball. The rul­ing may be tech­ni­cal­ly de­fen­si­ble and cor­rect un­der the law as writ­ten. Yet it can still dam­age le­git­i­ma­cy if the sport ap­pears to be gov­erned by in­vis­i­ble tech­ni­cal­i­ties rather than by pro­por­tion­ate, in­tel­li­gi­ble jus­tice.

Com­pound­ing

The is­sue is not my nos­tal­gia for the old ways but rather sys­tem de­sign. In fi­nance, we know that a risk mod­el can re­duce one kind of er­ror while cre­at­ing an­oth­er. A bank may adopt a mod­el to con­trol loss­es. But if every in­sti­tu­tion us­es sim­i­lar risk trig­gers, the mod­el can force them all to sell at once, am­pli­fy­ing the cri­sis it was de­signed to con­tain. The con­trol mech­a­nism be­comes part of the in­sta­bil­i­ty. Foot­ball faces a sim­i­lar prob­lem. VAR re­duces some ob­vi­ous ref­er­ee­ing er­rors, but it can cre­ate sec­ond or­der er­rors: se­lec­tive scruti­ny, in­con­sis­tent thresh­olds, over­con­fi­dence in mea­sure­ment and con­fu­sion be­tween tech­ni­cal de­tec­tion and sport­ing mean­ing.

It gets worse be­cause, the in­con­sis­ten­cy across match­es may be the great­est threat to le­git­i­ma­cy. In a tour­na­ment VAR is a hu­man-de­signed net­work of ex­pand­ed of­fi­cials op­er­at­ing un­der sub­jec­tive judg­ment laden thresh­olds. One ref­er­ee al­lows phys­i­cal con­tact; an­oth­er pun­ish­es it. One VAR team sees a clear and ob­vi­ous er­ror; an­oth­er sees a de­ci­sion with­in the ref­er­ee’s ac­cept­able range. One of­fi­cial in­vites an on-field re­view; an­oth­er silent­ly checks and moves on. These are all lay­ers of sub­jec­tiv­i­ty and dif­fer­ent match teams comes up with dif­fer­ent out­comes.

The tech­nol­o­gy has not elim­i­nat­ed sub­jec­tiv­i­ty. It has com­pound­ed and re­dis­trib­uted it. Be­fore VAR, the main sub­jec­tive ac­tor was the ref­er­ee. Now sub­jec­tiv­i­ty ex­ists in the orig­i­nal de­ci­sion, the VAR’s in­ter­ven­tion thresh­old, the choice of re­play an­gle, the use of slow mo­tion or re­al time, the de­f­i­n­i­tion of the at­tack­ing phase, the in­ter­pre­ta­tion of con­tact, the cal­i­bra­tion of the tech­nol­o­gy it­self and the fi­nal judg­ment af­ter re­view.

This cre­ates path de­pen­dence. The same penal­ty box con­tact can pro­duce op­po­site out­comes in dif­fer­ent match­es. If the ref­er­ee gives a penal­ty, VAR may de­cide there is enough con­tact for the de­ci­sion to stand. If an­oth­er ref­er­ee waves play on for the same con­tact, VAR may de­cide the non award is not clear­ly wrong. In both cas­es the pro­to­col can be fol­lowed.

Across a tour­na­ment, the per­mu­ta­tions mul­ti­ply: dif­fer­ent ref­er­ees, as­sis­tants, VARs, AVARs, re­play op­er­a­tors, cam­era an­gles, tour­na­ment styles, rep­u­ta­tion­al pres­sures and match con­texts. The re­sult is de­ci­sion vari­ance hid­den be­neath the lan­guage of tech­no­log­i­cal cer­tain­ty. VAR in its present form has sim­ply re­dis­trib­uted and com­pound­ed the prob­lem it was in­tend­ed to solve.

It is flawed to see VAR as tech­no­log­i­cal ob­jec­tiv­i­ty and ac­cu­ra­cy. It is a hu­man de­signed sys­tem with more lay­ers of sub­jec­tiv­i­ty than be­fore.

In com­plex adap­tive sys­tems, more da­ta and tech­nol­o­gy does not au­to­mat­i­cal­ly lead to bet­ter out­comes. More pre­ci­sion does not au­to­mat­i­cal­ly mean more fair­ness. A slow mo­tion re­play can re­veal truth, but it can al­so dis­tort judg­ment if used se­lec­tive­ly and with­out pro­por­tion.

Stock mar­kets taught us that long ago. Foot­ball needs to learn this les­son now.

Ian Nar­ine is a fi­nan­cial con­sul­tant who took ex­tra time to write this col­umn. Please send your com­ments to ian@ian­nar­ine.com