January 2025 – ZA Economist

No place to hide. The SA economy is in jeopardy for want of essential services.

by Brian Kantor and David Holland

We hear frequent complaints from state-owned enterprises (SOEs) that they are struggling to meet their debt obligations and need to be recapitalised. Yet the SOEs are simply not generating enough return on capital invested (ROIC) to justify allocating more capital to them.

Let’s take Transnet as an example. Transnet’s return on invested capital (ROIC) was 1.0% in 2024, which is lower than the miserable ROIC of 1.5% posted in 2023. The cost of debt on long-term South African government debt is 11%. This is the bare minimum cost of capital Transnet should meet. Transnet is destroying over 30 billion rand in economic profit every year when the difference between its return on capital and its opportunity cost is considered. A terrible unaffordable waste.

There are two key value drivers that feed into ROIC: operating margin and asset turns. To improve the operating margin, Transnet needs to increase revenue AND bring down operating expenses. To meet an 11% cost of capital, Transnet needs to generate 47 billion rand in annual operating profit on its invested capital. Its profit from operations was 4.3 billion in 2024 (operating profit is before interest expenses). Assets need to be run more efficiently or sold to private parties that can do so.

The large SA state owned enterprises, Eskom and Transnet, among other SOEs including the Post Office and the trading divisions of most municipalities have not remotely met the requirement for an investable credit rating. They have not been able to contain their costs well enough to cover their interest bills let alone their true opportunity cost of capital. Nor have they properly maintained their Plant and Equipment. Fundamentally they have failed because the bottom line, that is an adequate return on invested capital, has had little influence on their behaviour. Without many eye-watering billions of maintenance capex, their ability to continue to supply essential services, indispensable for economic growth, is severely compromised.

The SA government now recognises that the only practical way to secure the supply of these essential services is to raise large investments of private capital. Of which there is no shortage, provided the terms are regarded as acceptable.

And it would only be a private company, with a bottom line that requires a risk-adjusted return on capital, that could realistically hope to control costs and preserve and improve the capital stock, to reliably supply essential services in the future. Any imaginary mythically reformed SOE –capable of containing costs and operating efficiently – but without the right private incentives- is not a viable option. We’ve been down this track too many times.

The agreed terms for private sector engagement would have to recognise the true market value of the plant (PPE) taken over by the new operator of a network or part of it. Given years of neglect, existing assets might have very little value to any new operator. The potential market value could be exchanged for equity held by the government in the new privately controlled suppliers.

The private operator would have to commit the extra capital required to guarantee a flow of essential services over the long run. And such extra capex, would have to be recognised in the tariffs and or charges made for what could be valuable access to the networks.

Complicated calculations or regulations of appropriate charges or tariffs do not have to be a barrier to private entry. Any private company that won a tender to enter a network, with a degree of monopoly power, would be highly conscious of the impact of the prices it could charge on demand, sensitive to degrees of capacity utilisation and so its bottom line. And would be very considerate of its growth prospects. Becoming a reliable supplier at a competitive price with market-related returns would be a valuable outcome for all stakeholders.

The purpose of any private- public partnership or initiative is a simple one. That is to secure the future supply of essential services for SA users without which SA incomes and output must stagnate or decline. This all-important objective, viewed realistically, must be top of officials’ minds engaged in any negotiation with potential private partners.

Some Comments on how AI impacts the actions of the SARB

By Graham Barr

AI uses human actions in the realm of decision making; making judgements. So-called machine learning is central to AI. Optimal human decision making often follows a well determinable sequence in the form of “If this happens do this, otherwise do that” etc etc Machines can learn this best sequence of actions to mimic a human with the best judgement in that situation. Of course, some humans are intelligent to the point that they are known to solve particularly hard problems and are known for their ability to think “out of the box”; those that might solve problems that have never been solved before and thus no solution is recorded as existing to the problem at hand. Notwithstanding this unusual case, typical AI applications are developed with machines that are “trained” to replicate the best solution to a potential problem available. One might surmise that we could soon reach the point where machines could out-think the most creative human thinkers and prove new mathematical theorems and push the boundaries of knowledge. Will AI be able to write brilliant novels and think of ways and (manufacturing processes) to produce faster computer chips? Or think of ways to genetically modify plants so that food production eradicates hunger using a relatively small area of land. At some point, but we aren’t there yet.

AI is sometimes mixed up, I think, with “raw” computer power. Computer power is the straight-forward speed of a computer chip and then (separately) the ever-lower price of computer storage. Computer power is about the speed at which one can do an operation; whether it takes 1 hour or 1 second. And this leads on to whether a computer can process huge amounts of data in a useful amount of time – is a problem solved in an hour worthless, whereas a problem solved in a second is very useful. The rise in computer power over the years has been phenomenal (Moore’s Law) A terabyte of data might be processed in 30 seconds now but have taken 150 years to process 30 years ago. This raw data processing power which affects the time something takes to process is often a key requirement of AI; but should not be confused with AI itself.

Turning to consider clear examples of an AI centred approach which are useful, and not dependent on any significant computer power. The replacement of tellers at supermarkets would constitute a first-stage form of AI. This constitutes replacing a task not requiring much creativity; requiring a fixed pattern of behaviour in a deterministic (reliably set) way – this would be a low-level AI application. This application works quite well through product scanning, but generally requires an overseeing assistant to sort out problems. Bank bots answering questions through text responses to typed questions, exist currently, but appear to be extremely inadequate at answering anything except the simplest of questions.

AI could be used to direct drones to fly over fields of plated crops and look for disease; then if found go and collect an appropriate spray and fly out again and spray the afflicted crop. Or those with a sickness could be checked in through an AI-driven

diagnostic process and then have medication administered. Where appropriate, they may even be operated upon by an AI driven robot.

The “box” of recorded human intelligence is still circumscribed. AI can be taught (machine learning) to write simple essays by reaping bits of seemingly required knowledge from the internet. However, It can’t go further out the box than it has been trained to do.

Forecasting.

I think economic forecasting has always challenged economists. Interestingly, It is not obvious that over the period of the last 30 years, in which computer power has improved by a factor of 1m, that economic forecasting has improved at all. This is in contrast to weather forecasting which is decidedly more scientifically deterministic. Forecasts are determined according to the (short-term, at least) predictable changes in pressure, relative humidity, temperature etc Such forecasts have improved greatly over the same period with increased computer power rather than any clear use of AI.

Economic forecasting is more difficult because it requires capturing both rational and (ex post anyway) seemingly irrational actions of humans. An important point is that humans can perform actions which are irrational (not optimal) because of a short-term emotional response. Machines, in contrast, currently can only perform rationally. Currently, that it. It is conceivable in the future that machines could be trained to capture the emotional reaction to any event at all for each individual person on the planet.

Back to economic forecasting. If one Googled “AI for economic forecasting”, one gets a list of statistical procedures that have been around for 50 years. ARIMA(Box-Jenkins), Trend-Seasonal decomposition forecasting, scenario-based econometric model forecasting etc These processes have little to do with AI as discussed. They do benefit a bit from increased computer power, mainly that they are quicker to estimate and allow more models to be considered within a period of time. But this is not material. In fact, economic-type forecasts have not improved much over 50 years. The most obvious example of this is the fact that “pollsters” got nowhere near to forecasting the results of the recent American election. If anything, one could say that they did a worst forecasting job than has ever been done.

The governor of the Reserve Bank Lesetja Kganyago may be mixing up AI with raw computer power. Increased computer power will allow his scenario-based models to be estimated with many more variables with many more assumptions, giving many more possible forecasts over many more different time periods. This will make his task rather difficult as he tries to decide which possible future set of so-called exogenous variables might pertain in the future. There is no obvious reason that he will do any better than the seemingly-unfathomly-bad forecasts of the American election pollsters.

Is AI able to tell us how to fine tune our SARB economic models so that they will produce better economic forecasts. Probably not. Could one CURRENTLY put an AI powered machine to better manage the monetary policy of the SA Reserve Bank. Unfortunately, it is quite likely.

A tale of two cities – it is the best and the worst of times for home owners.

The differences in the value of residential property in Cape Twon and Gauteng – and in Durban and the upper market suburbs of all the other SA cities are most striking. I learn anecdotally of a modern multi-bedroomed home in a gated estate near Johannesburg, largely grid free, that is municipality free, that has been on the market for six months for an asking price of not much more than three million rand with no takers. The owner wants to move to Cape Town as do so many and the market is flooded with like homes. Palaces that would take R10m or more to build. The same 3 million would have difficulty in securing a well appointed two-bed roomed apartment in Sea Point.

The differences in property outcomes are easily explained. In the Cape property owners in the towns and suburbs get reasonable value for the wealth taxes and service charges they are forced to pay. This adds to rather than subtracts from property values. Unlike the abject failures of service delivery almost everywhere else that destroys demand for homes and valuations.

The sad news is that the hopes of many have been dashed. Those who have responsibly been saving, paying off their mortgages each month expecting that the growing equity in their homes would help fund their retirements when they downsized and cashed in. The expectation now is that the value of their homes has not and will not keep up with inflation.

I am informed by a leading property appraiser that on average commercial real estate in Cape Town is expected to increase in capital value by 6.5% p.a. – ahead of inflation expected – while equivalent property in Johannesburg is expected to gain but 3% p.a. Furthermore, the so-called capitalisation rate at which future rental incomes are discounted is lower for Cape Town property that a similar building in Johannesburg. A value adding 1.25% margin in favour of Cape Town discount rates is estimated.

Recent property transactions confirm this difference. The Table Bay Retail Mall was sold for an initial yield of 7.75% to Hyprop. The Mall of the South in Johannesburg was sold to RMB at a significantly higher initial rental yield of 9.5%. For landlords of residential property in Cape Town initial rental yields (net rents/Market Value) range from about 3% in the most favoured suburbs to about 5-6 percent where there is less competition. All consistent with expensive real estate.

It should be assumed that the total return expected by investors in real estate will be similar everywhere in SA. Flows of mobile capital make it so. But total returns come in two parts- an initial yield – in dividends or net rental income and in capital growth. Thus the more growth expected in net income the lower will be the initial yield for homes of similar quality. For cash strapped migrants relocating to the Cape it may seem easier to rent rather than buy- with expected capital growth conceded to the investor.

The good news for our cousins in the North is that their cost of living is dramatically lower than ours in Cape Town. The bigger and better the home the more consumption power it delivers and such delivery comes much less expensively in Joburg.

The opportunity to live in an inexpensive palace rather than a crowded bed-sitter for the same modest outlay, either in rentals paid or sacrificed by owner- occupiers, should surely be an attraction not to migrate. And enough of a saving to help owner-occupiers go off grid and behind gates. And to run a four by four to negotiate the potholes when off to their bushveld retreats.

My New Year advice therefore is to buy more of what is cheap. To house up and take advantage of the bargain basement offers. And hope that all the bad news about service delivery is in the already low prices. The best performing asset class in SA in 2024 has been listed property. Residential property might just enjoy a similar rebound should service delivery unexpectedly improve rather than deteriorate further. (see below)

Total Return To end Dec 2024. Property Vs The All Share Index

Source; Bloomberg and Investec Wealth & Investment

Data dependence – not forecast dependant

Reserve Bank Governor Lesetja Kanyago is hoping that AI will help speed up the updates of its economic forecasts. At a recent media briefing he remarked that “From the time we make assumptions to the time they go and generate a forecast, it could be a couple of weeks,” “But there is emerging evidence that says that AI could help us speed up the forecasting process. So, we might be able to generate forecasts quicker, and that speeds up our own internal processes.”One of the things … we have to look at is to what extent this technology can be utilised to speed up processes and make central banks very efficient.” News 24.

Which reveals an important reality. The forecasts of the SA or indeed any economy will only be as good as the accuracy of the key assumptions made that drive the model. The hoped for internal consistency of the forecasting model is not sufficient to the purpose. Accurate assumptions about the key drivers of the model, the exogenous variables as they are described in model building, are all important for accurate forecasts- and hence may need to be revised frequently

It is always necessary for forecasters to play catch up with changes in the environment. It makes any central bank with influence on a domestic economy data dependent- not forecast dependent. The term data dependent is a constant refrain of the FED and the SARB and inevitable given the forecasting failures.

Not only the SARB but every analyst everywhere will have turned to AI to help their forecasts. And every SA economist will be looking to AI to help predict the interest rate actions, or should we say reactions of the SARB.

They will know, as AI will have digested, that interest rate settings in SA will be little influenced by the current state of the domestic economy. As has been demonstrated again very recently. With lower inflation has come news of a decline in Q3 GDP. Meaning a still very high after inflation Repo rate accompanying a very weak economy and a reluctance to lower interest rates more aggressively. No model of the SA economy would deny that real interest rates influence expenditure, inflation and GDP. But inflation will also depend on what is happening on the supply side of the economy – particularly to the foreign exchange value of the ZAR and the prices attached to imports and exports. One doubts that AI will make the USD/ZAR exchange rate and hence inflation highly predictable.

If past performance is anything to go by a concern for the state of the economy and the supply side forces acting on inflation will not much influence the SARB. It will defend its narrow mandate – to control inflation- and therefore to have to fight not only inflation but inflation expectations that are thought to feed back to inflation itself. Any model that forecasts inflation using inflation expectations as a driver, must have an equation or two to explain inflation expectations. The influence of the ZAR on inflation and hence inflation expected would have to play their part.

Judged by the interest rate spread between US and RSA five year bonds, the ZAR was predicted last week to lose an average 4.7% p.a. over the next five years. Hardly a backdrop to less inflation expected over the longer run.

Data dependence must remain the SARB mantra given the unpredictability of the ZAR and its weak bias. Nor should concerns about inflation expected, so difficult to influence, be the fall-back position of the SARB to justify depressingly high interest rates. When the data clearly indicates that aggregate demand is falling below potential supply and the threat to inflation is not domestic in origin, the case for lower interest rates should not be denied. Matching domestic demands with domestic supplies is as much as the SARB could contribute to the control of inflation. Inflationary expectations, not inflation, are largely impervious to the state of the SA economy – as AI would confirm