In Business Model, Internet Access, Mobile, News, Spectrum Futures Conference

Artificial intelligence (machine learning) can make a key difference in forecasting telecom industry market share trends and performance, says James Sullivan, J.P. Morgan head of Asia equity research (except for Japan).

Beyond that, non-traditional data sets can shed light on “value,” as quantified by “quality” and “price,” Sullivan says.

Consider the key matter of market share. “When overall industry revenue growth is stagnant, as it is with most telco markets, market share movements become the determining factor of an operator’s top-line performance,” says James Sullivan, J.P. Morgan head of Asia equity research (except for Japan).

The implication: if your forecasting models are better at detecting changes in market share, your forecasts for revenue also will be more accurate.

That is important as forecasting changes in market share trends and anticipating trend breakage has historically been challenging for the Street,” says Sullivan.

He points to the case of forecasts for AIS from 2010 to 2016. AIS posted strong market share improvements from 2010 to 2012, which resulted in revenue significantly beating analysts’ forecasts. As analysts extrapolate from past performance to model future performance, they missed a trend change.

The same thing happened with trends changed again between 2014 and 2016, when competitor True deployed its 3G/4G networks, resulting in less AIS revenue than history suggested would be the case, from 2014 to 2016.

“For now at least, the uses of advanced analytical techniques, such as machine learning, are mainly applicable to the gathering and processing of alternative data sets,” says Sullivan.

“In our view, we can take a significant step towards solving long run market share and capex forecasts for telcos if we can understand and obtain the following data sets:

1) What is relative network quality by region, in terms of download/upload speeds but also network availability by technology?

2) What is current data pricing for each operator in each country analyzed, updated constantly?

“These statistics, taken together, define relative value and assist in forecasting pricing power by operator, market share shifts, incremental capex, and therefore incremental opex and margins,” argues Sullivan.

“In our view, capex is dependent on demand for data and data usage and current utilization/coverage of an operator’s network,” says Sullivan. “Changes in data pricing can signal how demand for data will evolve, while relative network quality can be a signal for utilization rates.”

The point is that data pricing and network quality are the key data sets required to forecast capex.

Networks “are not commoditized and therefore value is a function of price and quality.” So it is “value” that ultimately drives usage, capex and opex.

Sullivan will be speaking at the Spectrum Futures conference.

Start typing and press Enter to search