Can Ange Postecoglou's high-risk Celtic style work long-term? - James Dailey

But will it always be champagne and caviar with Ange Ball?

With my experience in analytics over the years including investment markets, I know concepts like sequencing and volatility of returns are significant considerations for investors. I use many of the same concepts when I am analysing football metrics and performance levels, as well as trying to forecast and project future performance and potential risks.

Let’s start with a simple two-game comparison:

There is nothing particularly special about these two games. I selected them due to the nature of their xG for Celtic versus the opposition. I ran the games through a Monte Carlo simulation engine, and here were the probabilities for each game:

As people involved in football analytics usually say, single-game xG is not terribly useful in explaining actual results. In such a low scoring sport, single-game variance is so significant as to make results diverge from underlying performance metrics regularly. This divergence between results and performance levels is one form of performance volatility.

While the difference in xG between the two games was just 0.01, and the respective win, draw, and loss probabilities seem pretty close, small differences can add up over the course of a 38-game season.

The probabilities are part of a normal distribution, which includes a measure called standard deviation (meaning that about 66 per cent of the simulation outcomes fell within one standard deviation, 95 per cent within two, and 99.7 per cent within three).

For the two games, one standard deviation for Celtic’s level of xG output were 0.95 and 0.98 and 0.70 and 0.84 for their opponents. Given the level of xG, one can see how large that variance is.  

These differences introduce an important concept relative to volatility in football results; conceding a higher amount of goal scoring chances increases the risk of dropping points even when the margin of outperformance remains consistent.

READ MORE: The Ange Postecoglou Celtic philosophy that hasn't changed since day one

At the level of conceding 0.93 xG in the Hearts game, the standard deviation was 0.70, or about 17% lower than when conceding 1.45 xG in the Motherwell game.

Of course, this is an overly simple example using just two games. Risks and volatility tend to compound as complexity increases, which is why in investing most people look to diversify in order to try and reduce these factors over time. If we extend our simulation to a 38-game season as if these two games were played 38 consecutive times, we can look at expected points as a gauge to try and assess potential volatility. Again, this is an exercise to look just at the statistical measures.

We see the impact of the higher volatility over the course of 38 games is reduced in this simulation from 17 per cent in the single game, down to about a 3.5 per cent difference at one standard deviation. However, the risks remain in the tail. As I’ve shown in this table, the difference in expected points on both the right and left tails for three standard deviation level.

Risks can layer in a way that compounds and creates 'fat tails', as Celtic unfortunately experienced last season. I wrote about that potential risk on Twitter in August last year, a Celtic lost to Ferencvaros, retained players who wanted to leave, went to Dubai, recruited players who did not suit the manager’s preferred playing system, and then retained a manager who seemed incapable of making selection and tactical choices to diffuse risks rather than compound them.

On the flip side of that coin, Rangers enjoyed a lot of positive variance last season, which I would argue thrust them into their right tail. Following the cup loss to St Mirren in December, Rangers’ underlying performance metrics dropped to levels where the probability of losing or drawing games increased to significant levels.

Note how many games were around that dotted line, where the zero bound for xG Difference is shown. I count 12 games where the difference, even considering penalties, was closes to a coin flip.

In contrast, here was Celtic for last season:

Not as dissimilar as one may expect given the ultimate 25-point gap. Here were last season’s differentials for goals and xG with and without penalties included. Rangers outperformed significantly on both metrics.

While significant variance was likely part of the outcome, I do believe their system-driven approach to defensive tactics, which limited opposition chance creation well, was also an important variable. Essentially, they amplified the benefits of having positive variance by enjoying it while limiting how often “bad luck” could emerge.

This brings me back to my two-game example and Ange Ball. All things being equal, an all-out attacking style may be thrilling and offer a compelling entertainment experience, but it also probably increases normal risks inherent to football a season. 

During Ange Postecoglou’s tenure at Yokohama F. Marinos, his title-winning team won the 2019 league with 70 points, outperforming their expected points of 57.4, third-highest in the league. The next season, his side finished ninth on 47 points, despite expected points again being third highest at 58.2.

With what may be an inherently volatile playing style, I believe this increases the importance for Celtic to manage other knowable risks, such as injuries and recruiting players who fit into Ange’s system. I’ll also be hoping for some good, old-fashioned luck. After last season, heaven knows we could be due some. Rangers losing to Dundee United in a game where they had a negative xG Difference was a good start!