Forecasting Inflation: Can Machines Outperform Economists?

Dec 4, 2023

On the cusp of a new year, the question top of mind for many governments, businesses and individuals must be “What will economic growth and inflation be in 2024?”  

After all, 2022-2023 may well be remembered as the years that central banks and economists made major missteps in forecasting inflation as they underestimated the scale and severity of rising prices, as well as the subsequent slow rate of recovery.

What resulted from this was global price surges and financial instability. Businesses and households saw price increases for essential goods while workers turned to industrial action to negotiate for wage adjustments. Global inflationary pressures persisted, despite best efforts of governments to manage monetary and fiscal stimulus.  As they responded to this cost-of-living crisis with a slew of aggressive interest rate hikes to convince consumers and investors of their commitment to ensuring low inflation and sustainable growth, central banks’ actions were met with widespread scepticism and discontent.

 

Improving Forecasting in an Unpredictable World

For much of the 21st century, central banks and economic experts have played a vital role in forecasting and managing global inflation to support long-term economic growth. Longstanding methods of forecasting focused mainly on several key macroeconomic variables to predict inflation, using a balance of monetary, wage and employment variables.

Our post-pandemic world has, however, posed a new set of challenges for central banks and other institutions that trade on inflation and economic forecasts. Today’s networked digital global economy is characterised by fast-changing, unpredictable and complex operating environments. Global and national events – ranging from geopolitical tensions to supply chain disruptions– are driving structural changes across key markets and industries.

It is clear that underpinning the global economy today are highly interconnected market and economic variables that share very complex relationships. It is against this backdrop of volatility that central banks are continuing to grapple with not only forecasting inflation, but using their forecasts to set policies that aim to influence price pressures looking years ahead.

So, whilst forecasts can never be expected to be 100% accurate, the goal of delivering more precise, useful and insightful predictions about economic and inflation trends remains. In light of the 2022-2023 missed forecasts, the question remains if better strategies exist to support inflation forecasting and hence economic stability.

 

Achieving Clarity in the Age of Big Data

One of the strategies for financial institutions to gain much-needed clarity to navigate the complex global business landscape lies in leveraging new technologies, in particular machine learning and big data. A major competitive advantage of a machine learning and data-driven approach is the elimination of guesswork and biases that may inadvertently derail forecasting accuracy. Coupled with the power of machine learning, which is subject to continuous training to capture and model variable relationships more cohesively, financial institutions and other organisations can be better equipped with data to deliver more robust inflation forecasts.

To this end, a unique cutting-edge platform to support this vision has been developed by Turnleaf Analytics, which applies machine learning to an extensive collection of over 10000 data variables – comprising traditional datasets that are augmented with alternative data (See Box A – Data Types). Sourced from central banks, official statistics organisations and various data vendors, this collective dataset covers 32 developed and emerging markets, representing more than 95% of investible markets (See Box B – List of Countries).  Careful processing of raw data – from checking for outliers to data cleaning and organization into datasets – sets the stage for effective application of machine learning and data mining analytics.

 

Gaining global inflation insights to minimise risk and maximise opportunities

Recognising that financial institutions require inflation and macroeconomic forecasts for very different purposes, from trading FX to setting pricing and interest rates, a key focus for Turnleaf Analytics is to apply a purely quantitative process to wide-ranging macroeconomic indicators and alternative data, in order to create a multitude of price forecasts for different markets and for different time horizons.

Today, more than 100 economic inflation prediction models are being operated which generate forecasts up to 12 months in the future. Most recently, Turnleaf Analytics has rolled out higher-frequency updates for 20 markets. To provide additional insights into where inflationary trends and market factors intersect, we track model forecasts (as well as consensus and realized inflation in time) together with an attribution analysis.

Additionally, recognising that microtrends and sentiments can contribute to market perceptions on inflation trends, Turnleaf’s comprehensive system includes a Fed Communications Index, which ingest text data generated by the Fed and applies natural language processing to create an independent and quantitative viewpoint on the sentiment of what the Fed is communicating to the markets (see Turnleaf Fed Communications Index). A Sentiment Dashboard is also available which draws from global media sources daily to map global perspectives and societal sentiment around the latest macroeconomic, economic, inflation and geopolitical news and trends (see Turnleaf Sentiment Dashboard).

This multi-pronged strategy to leverage data to gain global inflation insights has been tracked and shown to deliver positive results, ranging from more accurate forecasts to increased trading revenue. Highlights include:

  • More than 3000 tracked forecasts completed since May 2022 have shown an accuracy rate of 62%, beating benchmarks’ performance across all countries and time tenors.
  • Accurate prediction of inflation direction 84% of the time across all tenors (1M-12M)
  • Mean absolute error 1.03%, compared to that of the benchmarks of 1.25% (as of 31 Oct 2023)
  • In bond futures trading, Turnleaf Analytics demonstrated in live data a Sharpe ratio of 1.3 and annualized returns of 9.3% since 2018, compared with a trend-following strategy that with values of 0.6 and 4.3% respectively

 

Conclusion

Navigating in the post-pandemic economy, central banks, governments and businesses are at an inflection point as they look for new ways to stabilise inflationary pressures without stifling growth. Initial results produced by new technologies, in particular machine learning algorithms applied to alternative data sources, have shown promise by outperforming economists using traditional forecasting methods on certain benchmarks. Whilst not a panacea for inflation forecasting, a data-driven approach will lead the way in combining the best of different approaches for enhanced inflation forecasting accuracy. With their experience and insights, experts gain an added advantage by incorporating new technologies for direction-setting and strategic policy development to manage inflation and build a stronger global economy.

 

Alexander Denev, co-founder Turnleaf Analytics.       Contact: Alexander@turnleafanalytics.com