Inspired by Gregory Zuckerman’s book on Jim Simmons I thought that I’d try some Markov Chain analysis on company announcements. Although it is not entirely clear what Simmons does to generate consistently high returns year after year, it seems to be something to do with Markov Chains, or hidden Markov Processes. Markov chains assume the outcome of one event depends directly on the preceding event (and only on it). For instance, in the English language that the letter “t” is often followed by and “h”. “For” is followed by “the” roughly 4x more often than “example”. Because it was used in cryptography, as well as Simmons, Claude Shannon knew a thing or two about Markov chains.
Selling brands to Africa
I was curious to apply this type of analysis to PZ Cussons financial announcements, because PZC is an example of a multi-bagger that has lost its way. The PZ of the title stands for Patterson and Zochonis, a Scot and a Greek who founded the company in the 19th Century to export textiles from the UK to Nigeria. In 1975, they bought Cussons Group to grow its original markets in Africa to cover the Far East, the Middle East and Eastern Europe. In May 2002 the company changed its name from Patterson Zochonis to PZ Cussons, to create a global brand. Their brands include Imperial Leather (soap), Original Source (shower gel) San Tropez (self tan), Rafferty’s Garden (baby food), Fudge (hair styling).
When PZ listed on the UK stock exchange in the 1950s, it was worth just over £1m versus £1.7billion in 2010. Decade after decade of performance generated almost a 1000 bagger. But the last decade has been less impressive, with the market cap and share price halving to £800m and 200p respectively. Add to this that Unilever, PZ Cusson’s larger competitor has continued to multi bag this decade – so it would be great to know if there were any warning signs that investors could have identified beforehand.
Returns are rarely average
Or put another way: since the start of 1999 PZ Cussons has compounded at 9% (not including dividends) close to the long term stock market average. But that average over two decades hides a changing performance, the first decade the share price compounded at 22% per annum, falling to negative -7% in the second decade. In a previous post, I talked about this study looking at the average return of the index hiding a pareto distribution: The top-performing 1.3% of firms account for the $US 44.7 trillion in global stock market wealth creation from 1990 to 2018. Outside the US, less than one percent of firms account for the $US 16.0 trillion in net wealth creation.
I think the same goes for the time average performance of multi bagging stocks, probably the best few years contribute disproportionately to long term returns. PZ Cussons best year was 2001, when it rose 154%, and it’s worse year was 2018 when it fell by 33%. That 2018 performance should be seen in the light of the fact that as early as 2012 PZ Cussons was warning about problems in Nigeria, what should have been a key market for growth and opportunity. So those long term time averages hide a much greater variation in returns, which could create an opportunity if we can use text analysis to identify when to sell before things start to go wrong.
But before I do a lot of text analysis, one thing that I was curious about was whether there had been more positive or negative announcements in the noughties when the share price 10 bagged from below 40p to 400p or in the most recent 2010-2020 decade. I took 94 price sensitive announcements going back to 1999, and labelled them as either positive (if the share price increased to the next announcement) or negative (if the share price fell to the next announcement). This is rather subjective, I haven’t counted directors dealings or large shareholder buying and selling as containing information. Instead just trading statements, AGM statements, half and full year results. There’s obviously a grey area, for instance I did count the change of name announcement to PZ Cussons as price sensitive, which is debatable. But generally I think 4-5 announcements per year sounds about right. To give you some idea see the announcements for 2018. NB PZ Cussons has a 31st May year end.
|Analysis of PZ Cusson announcements 2018|
|Date||Announcement||Share price (p)||Reaction (%)||+ or – tive?|
|26-Sep-18||AGM Trading Statement||235||-11%||Negative|
The other thing to point out is that the frequency of announcements increased in the most recent decade. So of the 94 announcements, roughly 1/3 was in the first (ten bagger) decade, and 2/3 were in the second (when the share price halved) decade. That is, as the company’s performance worsened, its communication frequency increased.
It’s also worth noting that I can label announcements as positive or negative, but the percentage share price move makes a huge difference. For instance, the worst share price reaction in the first decade was a -13% fall following the Sept 2008 Interim Management Statement. Whereas since 2015 there have been 5 negative announcements with share price falls greater than -13%, including the half year results in January 2018, when the share price fell 26%.
|Time||Neg Neg||Pos Pos||Pos Neg||Neg Pos|
Markov chain analysis – I was interested if announcements were mean reverting (ie a positive announcement was likely to be followed by a negative announcement) or whether there were long streaks of positive and negative announcements. The bottom row of the table (which sums to 1) shows that over 20 years announcements were very evenly split between the 4 alternatives of Negative followed by a Negative, Positive followed by a Positive, Positive then Negative and Vice Versa. But in the first decade 1999-2009 a Positive announcement followed by another Positive announcement (column 2) was almost twice as likely as two negative announcements (33/18). The situation reversed in the most recent decade, with 2 negative announcements about 1.5 times more likely (28/19).
Regime change in Africa?
Quants like to use the phrase “regime change” to signify a shift from trending markets to mean reverting markets. This is only a metaphor, but I think it comes from Markov’s Model of the transition probabilities. One example given by Scot Page in his book is of countries progress towards democracy. Over the past 35 years the percentage of “free” (as categorised by Freedom House data) countries has been rising. If the trend continues by 2040 2/3 of the world’s countries will be categorised as “free”. Rather than this simple linear extrapolation of the trend, a Markov Model leads to a different prediction. To make predictions he sets the length of a time period to 5 years and looks at transition probabilities based on historic data. The Markov model makes less rosy predictions because it allows for “free” countries to transition back into tyranny – which seems a more realistic assumption given 20 century history, but also ancient Greek and Roman history. As more countries become free, the number of countries that might transition back into tyranny also increases – and the Markov model captures this.
Simmons has obviously done well out of the insights that Markov Models provide. For those who believe in Efficient Markets his success is rather hard to explain. But he needs highly liquid markets and has to trade frequently to profit from this. I’m more interested in using text analysis to help with a “buy and hold” strategy, emulating Claude Shannon. PZC (which I bought recently) is a stock that interests me because it has transitioned from long term multi bagger to disappointment. I’d like to understand what has gone wrong in the last few years, and whether there were any signs in the text that warned it was in for a tougher time. Or indeed, whether there are signals that the company has now turned the corner and is back to it’s old multi-bagging long term performance. I’m planning to return to that in another post.
- The Model Thinker – S Page https://www.amazon.co.uk/dp/0465094627/ref=cm_sw_r_tw_dp_U_x_yLZcEb42YDKY4