An Italian muddle

Markets may well be in for another Italian surprise. We do not make this claim lightly. In electoral forecasting, as in other areas, experience teaches you to be humble. Over time, you acquire rules of thumb that tend to be broadly applicable. And, if you do it long enough, you eventually tend to find an exception to each and every rule. Italy's looming parliamentary elections look like a case in point. On our reading of the admittedly ambiguous data, we believe the anti-establishment Five-Star movement may only be a normal polling error away from being able to block the formation of any realistic, broadly centrist, pro-European government.

On March 4th, Italians will elect new members to both the Chamber of Deputies and the Senate – together they make up the Italian parliament. This includes all 630 Deputy seats, as well as 315 out of 321 Senators.[[DISCLAIMER:Former Presidents are ex officio life senators. In addition, the incumbent President may appoint five life senators for "outstanding achievements in the social, scientific, artistic or literary fields." (Article 59 of Italy's constitution.) Currently, there are five appointed Senators for life, plus one former President, 92-year old Giorgio Napolitano. See: http://www.senato.it/3801 for further details.]] Under the new electoral system, both chambers will be elected through a mix of proportional representation via party lists (for approximately 63% of seats ^[1]) and first-past-the-post ^[2], single-member seats (the remaining roughly 37%). The rules still differ slightly. Notably, Italians need to be 25 years or older to vote for the Senate. But overall, the new system should ensure that both chambers will look very similar in terms of their party composition. This marks a big change. In the Italian system, the two houses perform identical functions. Previously, this and the at times very different party composition in the two houses contributed to gridlock.

Other tentative signs of progress include how the election is seen by financial-market participants. After 2016's market-moving surprises, more and more investors appear inclined to think in terms of probabilities, rather than certainties. As a general trend, the spread of probabilistic thinking is more than welcome. We have previously pointed out that the polling data was pretty ambiguous ahead of the Brexit referendum and the U.S. presidential election.^[3]

Which takes us back to Italy. As we noted last year, the constitutional referendum of December 2016 was by far that year's biggest polling disaster ( March 2017 European Election Special ). The far-reaching changes were rejected by a whopping 18% margin, more than 3 times the unadjusted polling average in the weeks leading up to the vote.

This was an early reminder that some of our favorite forecasting rules of thumb might not apply so well in the Italian context.

In particular, we are thinking of the following three rules:

• Be wary of consensus views in political betting markets or in financial markets. At the very least, you should be as cautious as you would be when trying to pick a winning stock in equity markets, by looking at broker recommendations rather than doing your own analysis. Chances are that your fellow amateur forecasters might be living in an echo chamber. They may be too far removed from actual voters to know any better than the pollsters.

• Instead, take a polling average. That way, some of the polling errors might cancel each other out. Use quality weightings if there is a solid basis for suspecting some polls or polling methods might work better than others. For example, you might assign a higher weighting to pollsters that have been comparatively accurate in the past.

• For example, you might be able to further improve forecasting accuracy by looking at economic fundamentals (e.g. a strong economy benefiting the outgoing government) or social-media indicators.

In general, these rules work quite well. And, broadly speaking, the publicly available polls paint a reasonably reassuring picture. On fairly unsophisticated quality-adjusted polling averages, the center-right coalition is ahead with about 37% of the vote.[[DISCLAIMER:We are using 31 recent surveys, overweighting the 6 historically more reliable ones. An official collection of the data by the Italian government is available at: http://www.sondaggipoliticoelettorali.it/ListaSondaggi.aspx?st=SONDAGGI For an easily available English language overview of Italian polling throughout the campaign, see https://en.wikipedia.org/wiki/Opinion_polling_for_the_Italian_general_election,_2018 ; ]] The anti-establishment Five-Star movement is at about 28%. The center-left coalition is at 27%; its main member, the ruling Democratic Party, led by former prime-minister Matteo Renzi, is at just over 22%. Among the smaller, unaffiliated parties, the only one likely to clear the 3% hurdle to get into parliament on its own is the left-wing Free and Equal alliance at 6% (itself a spin-off of the Democratic Party).^[4]

Under the new electoral system, former prime minister Silvio Berlusconi's center-right coalition would most probably need a vote share of at least 36% to get close to winning a majority in both houses. How close is impossible to say at this point, numerous forecasting models notwithstanding. Under the new electoral system, small differences in the regional distribution of the vote share could have a big impact, particularly in shifting the balance of power for key marginal first-past-the-post constituencies in southern and central Italy.

Methodologically, the problem is similar to translating national polls into seat forecasts for the British House of Commons and related to getting the Electoral College in a U.S. presidential election right. You would need to know how well the center-right does compared to the next runner-up in each single-member districts, taking into account both last-minute shifts and regional correlations. That is pretty hard, as long experience in both the UK and the U.S. shows.

It is much harder in Italy, however, because there is little directly relevant historic district data, combined with an already very fragmented political landscape. The center-right coalition is composed of four parties: Berlusconi's Forza Italia (about 16%), its right-wing allies from the Lega Nord (13%) and the far-right Brothers of Italy (5%), as well as the centrist Us with Italy (2%). That makes it even riskier to extrapolate across various regions, than to say what the performance of the Labour Party in England might mean north of the Scottish border. All that can reasonably be said is that the center-right might be able to secure a majority with as little as 36% or 37% of the vote – if, for example, the center-left underperforms in regions it has historically been strong in. Conversely, Berlusconi's coalition might fail to get a majority with as much as 42% or 43% of the vote, if, for example, Five-Star turns out to perform strongly and consolidates those opposed to the center-right.^[5]

Apart from that, the only safe prediction may be that even if you could trust the polling, it would probably take at least three to four additional electoral cycles worth of data before investors should put much confidence into seat forecasts.

As it turns out, trusting the polls is a lot less advisable in Italy than in other countries. The big polling miss at the 2016 referendum was only the latest in a long series. This hints at severe methodological flaws, perhaps related to declining response rates, notably in traditional telephone surveys.^[6] In 2013, for example, a simple polling average of the last 27 published polls had Five-Star at just over 15%. In the event, the party won 25.6%, an astonishing 10-point miss. Conversely, polls throughout the campaign overestimated the center-left Democratic Party by about 5%, compared to the actual results. Previous elections show a similar, but smaller bias, with the center-left tending to do better in the polls than at the ballot box.

Looking at the last three general elections of 2006, 2008 and 2013, polling accuracy appears to have gotten less reliable over time. A charitable interpretation for this would be to blame the 15-day blackout period ahead of voting. In Italy, for the final weeks of the campaign, polls may be conducted but the results may not be published. This leaves any forecast of national vote shares highly vulnerable to last-minute mood swings. The deteriorating polling accuracy in Italy as elsewhere could then be due to a large and growing share of the electorate deciding too late to be reflected in published polls. Indeed, the results of late polls conducted during the 2013 blackout period and published after the voting suggest slightly smaller, but still sizable misses, notably on Five-Star.

Disentangling such last-minute mood changes from other issues such as declining response rates could probably fill a PhD thesis. In fact, it already has. Using a pretty robust econometric approach on the polling data from the last three Italian elections, Graziella Castro from the University of Salford concludes that all three "have been characterized by a high percentage of polls classified as inaccurate." (p. 239) Moreover, polling errors have indeed grown larger, mostly, it would appear, due to plenty of dodgy polling, rather than any systematic last-minute movement in voter sentiment.[[DISCLAIMER:Graziella Castro (2015) " Polling and Political Behaviour: Explaining inaccuracy in Italian polling", http://usir.salford.ac.uk/36898/1/GC_Thesis_v515V.pdf ; as well as: Graziella Castro (2013) Polling and Accuracy Measure: Evidence from the Italian Case (April 3, 2013). 71st Annual Conference of Midwest Political Science Association, April 10-14, 2013, Chicago, Illinois. Available at Social Science Research Network (SSRN): https://ssrn.com/abstract=2327184]]

To put things differently, Italian pollsters have tended to err in the same direction. The better ones simply did so a little less aggressively than the rest. Based on this, a strong case could be made for simply disregarding a lot of the published polling data – and for being very wary indeed of probabilistic seat and coalition forecasts based on conventionally received polling wisdom. In the Italian case, taking an unweighted average of all polls can make things worse, rather than better. And, extrapolating from flawed polling data to seat projections is likely to compound the forecasting error.

This takes us to the last two rules of thumb of this note:

• Given the size of previous polling misses, this would suggest adding a margin of error of say 5% in both directions for the center-right coalition and the center-left coalition, and perhaps as much as 10% for Five-Star. That effectively amounts to saying that it is impossible to predict who will form the next government.

• Of course, Italian polls could prove broadly correct. We would find this rather surprising, but stranger things have happened on election days in recent years. And, if the polls got it broadly right, we would expect to see the same sorts of outcomes that other observers are confidently predicting. Mostly, it would be continuation of the status quo. Cyclical growth could be strengthened slightly, especially if the center-right eggs out a narrow victory. In the longer term, we fear that the absence of further reforms would keep the Italian growth potential very low. From a market perspective, the best outcome would be a coalition between Forza Italia and the Democratic Party, perhaps including some smaller, centrist parties. This would probably offer the best chances of at least not reversing recent structural reforms. Forza Italia certainly appears ready to break with its right-wing allies, if another opportunity to govern presents itself. In case of an indecisive election outcome, there is also a non-negligible chance of either new elections or a technocratic "national government". In our view, the former would be negative – resulting in more policy uncertainty and paralysis – while the latter could actually turn out to be more positive than some of the alternatives.

• Note that all these, however, rely on Berlusconi's Forza Italia, Renzi's Democratic Party and their more centrist allies doing reasonably well. That may indeed turn out to be the case. In general, you might expect economic momentum to modestly boost the current government and the Democratic Party. Recent events may further solidify the voting share of mainstream parties, including revelations that some Five-Star politicians failed to follow party rules in making contributions to a fund supporting small businesses.

• By contrast, it is hard to predict the impact of the rise in racial tensions and violent clashes following protests by both the extreme right and the extreme left. However, patterns in other countries show that rising immigration concerns tend to reinforce the election-day performance of anti-establishment parties. In Italy, this could mean Five-Star, the Lega Nord and the Brothers of Italy; Five-Star has taken an increasingly aggressive stance on stopping illegal migration. Certainly, the rise in immigration concerns suggested in the polling could easily outweigh any sense of gratitude voters might feel towards the ruling left-wing government due to stronger economic growth.^[7]

• In key Southern constituencies, an anti-incumbency mood might also work against the center-right, specifically the candidates of Us with Italy, the smallest and most centrist among Berlusconi's allies. Their single-seat district candidates are generally veteran politicians. Many of them are local office holders and former members of the Christian Democrat party, once the dominant force in Italian politics. Seat forecasts generally assume Us with Italy candidates will benefit from strong name recognition in their local districts. However, this could equally well work against the center-right, by making Us with Italy seem like part of the establishment in the eyes of voters dissatisfied with the status quo.

• Meanwhile on the left, the ready availability of the left-wing Free and Equal party and smaller fringe groups further to the left could hurt the Democratic Party throughout Italy. This may leave the combined seat share of the left even lower than most forecast models are assuming. At least for risk-management purposes, we would therefore suggest deviating from one final rule we usually tend to adhere to:

• In a recent Bloomberg survey among economists, the probability of a Five-Star-led government had fallen to an average of 8%. And this outcome was also universally seen as the worst possible outcome for financial markets.^[8] To us, this 8% probability appears way too low. Many observers, we suspect, wrongly extrapolate from Five-Star's weak performance in last June's municipal elections, as well as the strong showing of the center-right in November's election in Sicily. This is almost certainly misguided, given the low turnouts at these votes. Relatively new protest movements like Five-Star have to rely on energizing their sympathizers anew before each vote. Low-turn-out elections tend to have very little predictive power one way or the other. This is likely to be especially true for municipal elections, given that Five-Star still only has a relatively weak and uneven local presence in much of the country.^[9]

• In addition, we suspect that Italian pollsters, like pollsters elsewhere, may genuinely be struggling to correctly model and predict turnout among younger voters. That certainly appears to have been a key reason behind their polling miss in 2013. Unfortunately, very few Italian pollsters disclose sufficient data to guess how turnout is modelled and what sort of demographic mix of the voting electorate is assumed. For what it's worth, though, some of the seeming stability in the polls ahead of several recent votes, followed by surprising election-day outcomes, may reflect systemic over-weighting of older voters, more likely to answer the phone and much more likely to adhere to their traditional left and right allegiances.^[10]

• If so, the result could well be Five-Star outperforming recent polling averages – as far as it is possible to tell from published data, its support remains skewed towards younger voters. The same is also true on the right for the Lega Nord and the new, centrist, pro-European +Europe party. Such shifts could have a big impact on what sort of government will be feasible and what agenda it might follow. For example, +Europe would be able to form a far larger share of the center-left parliamentary grouping it if managed to clear the 3% hurdle.

• By taking the polling data at face value, many market participants may be underestimating the risk of Five-Star and, perhaps, the Lega Nord and the far-right Brothers of Italy doing better than expected. The same goes for the risk of markets getting spooked. Keep in mind that on our quality-adjusted polling average, these three already have a combined vote share of 46%. By contrast, the mooted "grand" coalition between the Democratic Party and Forza Italy is at just 39%. An anti-establishment majority of the Lega Nord, 5-Star and Brothers of Italy in at least one of the two houses of parliament is a very real possibility, even without much of a polling error. Both the Lega Nord and the Brothers of Italy will probably be overrepresented compared to their proportional vote share, due to their respective strength in Northern Italy and around Rome.^[11]

On election eve, Italy could look even more politically fragmented and less governable. Both Five-Star and the Lega Nord have lately tuned down their rhetoric on Italian membership in the Eurozone, but it remains anybody's guess how lasting this change will prove. A strong showing for either or both would make resolving some European issues, notably in relation to migration, much harder. This could spill over into other policy areas. Fears over European politics could return much sooner than many are expecting.

To be clear, this is not our base case. Therefore, we are overweight Italian government bonds. Their spread vs. Bunds is still well above our target. On Italian equities, we take comfort from positive signs on both the Italian economy and earnings growth, which we believe should actually come in above the Eurozone average. Of course, that presupposes that politics does not get in the way. For now, that remains quite a big "If."