Late Bloomers screen banks on commercial property recovery

Though a well-worn term in computer science circles, in the context of AI the phrase refers to the limits of large language models and the propensity of chatbots like ChatGPT to confidently spew out inaccurate, misleading, or logically flawed answers to prompts.

Feed a model questionable inputs – such as, for example, the unfiltered musings, unverified data and conflicting sources of the entire internet – and its quality will be reflected in the output.

The stock screens that appear in these pages may lack the technical sophistication of machine learning and generative AI. But in practical terms, they are simple processes made up of inputs, rules, and outputs. For our purposes, the inputs are a wide array of stock data and fundamentals, the rules are whatever we are trying to filter, and the outputs are a screen’s selections.

Like AI, they can misfire. This week’s screen, for example, is strong evidence that even if you use good data and put plenty of thought and reason into a set of rules, you can still end up with ‘garbage out’.

This is perhaps especially true of the Late Bloomers screen, which has been in service since 2014. Unlike many of the screens that appear in these pages, its methodology doesn’t involve analyst forecasts, meaning we cannot blame ‘garbage in’ – market estimates that prove overly optimistic – for its failings.

But fail it has. Although the screen broadly kept pace with the FTSE All-Share for the first seven years of its life, it has come badly unstuck in the last two, leaving it with an all-time headline total return of just 18 per cent over nine years, compared with 59 per cent from the UK market.

Although the screens that appear in these pages are designed as a source of idea generation rather than ready-made portfolios, the total return drops to a paltry 3 per cent once a notional 1.5 per cent annual charge to represent dealing costs is factored in. Even the pathetic basic savings accounts rates on offer for much of the past decade would have been a far less risky – and probably more profitable – option.

Last year’s selections, while promising recovery, proved a particularly weak bunch. Just one – car finance specialist S&U (SUS) – posted a positive return, while the group was weighed down heavily by the broad sell-off in commercial property.

Looking back to years past, this is hardly an aberration. With a few exceptions, which I would put down to the ordinary dispersion of returns, the Late Bloomers screen’s selections have consistently lagged the FTSE All-Share or failed to filter out stocks poised to recover from the falling knives. Its value focus has not delivered with anything approaching the right kind of consistency – shares with alpha – meaning it is missing a critical component of a reliable screen.

Why, then, has it failed to flower? Originally formulated to identify cyclical stocks that were late to enjoy the fruits of improving economic conditions, it has instead run into two big issues.

The first is that it is too restrictive. By insisting that stocks pass all its rules, its selections have dwindled in recent years, leading to an ugly concentration risk. As I argued last year, the methodology (outlined in full below) feels highly tuned or even reliant on a set of macroeconomic conditions for it to work.

Worse still, defining exactly what those conditions are can be a challenge. While few economists would argue that we are in what you might call a recovery right now, it’s never entirely clear when we are late in an economic cycle – at least in terms that can be broadly applied across industries.

Insofar as the screen does work in periods of genuine market recovery (such as in the teeth of the pandemic in May 2020, when its selections’ 52 per cent total return more than doubled the FTSE All-Share), it seems smarter to deploy the screen in a genuine market crash. It may have just misfired in the past two years because there has been little in the way of a genuine recovery.

The second, seemingly contradictory issue, is that the screen is very generous toward real estate and financial companies, resulting in a high proportion of stocks from each sector in the past couple of years.

For both sectors, the screen’s waiver on positive free cash flow feels a lot dicier than it probably looked in 2014. So too do the diluted tests on leverage.

To my mind, there is an even more glaring hole in the recovery test. By giving Reits a free pass on margins and returns on equity, the screen was initially designed to get around the sometimes-lumpy nature of property investors’ statutory profits. But in effect it makes a big assumption: that the underlying demand for property is either inelastic or growing, and therefore not something to worry about. While this might hold true for residential property, it feels like an outmoded call in the wake of the pandemic, given the uncertainty around commercial property values and tenant demand.

In a possibly last-ditch attempt to address these bugs and reduce the screen’s overall exposure to property stocks, I have come up with what feels like an awkward compromise, the details of which follow the methodology below.

The method

The Late Bloomers screen looks for shares that appear cheap against their own long-term valuation range, by looking at the valuation of a stock against the ultimate source of its earnings. This is based on the observation that stocks often rebound after profits have been hit. By looking at a company’s future earnings potential (rather than current earnings) investors can get a better read of the prospect of recovery, at least in theory.

The screen assumes the source of earnings for most companies is sales, while for property firms, housebuilders, and financials, it is net assets. Turning this into a valuation metric requires a bit of maths – or more specifically, a standardised statistical measurement called a Z-score – which is used to see how cheap or expensive the valuation looks compared with the 12-year history. If a full history does not exist, the reference is a minimum of six years.

From this approach, we get the natty acronym of the ZEUS ratio (which stands for the Z-score of earnings’ ultimate source). The screen looks for a ZEUS ratio of -1 or less, suggesting the valuation is toward the bottom of the historic range. The other tests used by the screen are there to see if the balance sheet looks in a serviceable state and whether the company has been more profitable in the past than it is now. 

The full criteria are:

■ Value: ZEUS ratio of -1 or less. 

■ Recovery potential: Real estate companies get a free pass on this test. For financials, return on equity needs to be at least one-third below its 10-year peak. For other sectors, operating margin at least a third below its 10-year peak.

■ Balance sheet: (i) For financials, equity representing at least 5 per cent of assets and return on assets of at least 1 per cent. This is a test suggested by the great Peter Lynch, the former star manager of Fidelity’s flagship Magellan fund. (ii) For utilities, which have very defensive earnings streams well suited to supporting high levels of debt, net debt to book value (gearing) of less than 150 per cent. (iii) For real estate companies, gearing of 75 per cent or less. (iv) For housebuilders, gearing of 25 per cent or less. (v) For all other sectors, net debt of 1.5 times cash profits or less.

■ Growth: Growth in 'earnings' ultimate source' (the EUS in ZEUS) over the past 12 months. 

■ Positive free cash flow: Free pass for financials and real estate.

(NB investment trusts are excluded from this screen, but not Reits)

This year, just three stocks passed all tests. Somewhat frustratingly, from the perspective of diversification, all three are Reits, meaning the screen has gone all-in on a recovery in commercial property. And while recent quarters have shown some signs that asset values may have started to bottom, it feels like another sign of the screen’s overly accommodating view of real estate.

To broaden the screen’s narrow stock and sector selection, I have weakened the recovery criteria to admit companies that pass just one of the recovery tests. In effect, this means admitting six more stocks whose returns on equity are at least a third below their decade peak, given no company that passed the remaining tests has seen a dip in margins of at least a third.

In doing so, I feel like I am moving away from the spirit of the screen to improve its output. Despite doing so, I may choose to discontinue it next year. Like every investing strategy, screening has its flaws. But the Late Bloomers screen no longer appears to be working.

Programming note: For the next two months, an extended period of leave means I won’t appear to be working much, either. Regular screens will appear from the end of June.