Five years ago, if you wanted to forecast the Aussie dollar, you’d look at commodity prices, read the RBA’s statement, check US yields, and maybe throw in some technical analysis. Pretty standard stuff. Now? Half the market’s running machine learning models that can process thousands of data points simultaneously, and the game’s changed in ways that are both fascinating and occasionally unsettling.

I’m not talking about the “AI will replace all traders” nonsense that tech evangelists love to spruik. That’s fantasy. But the tools available for currency analysis have genuinely evolved, and if you’re still forecasting AUD movements with purely traditional methods, you’re missing part of the picture.

What AI Actually Does Well

The strength of machine learning in forex isn’t prediction in the crystal ball sense. It’s pattern recognition at a scale humans can’t match. Feed a neural network 20 years of AUD/USD data alongside iron ore prices, Chinese PMI figures, RBA cash rates, and US employment numbers, and it’ll spot correlations you’d never find manually.

I’ve been testing a couple of these systems over the past year. The good ones don’t claim to predict the future. They flag when current conditions resemble historical patterns that preceded specific moves. It’s subtle but useful. For example, one model highlighted last November that AUD behaviour relative to CNY was tracking 2019 patterns—three weeks before we saw that sharp correction in December.

The Reserve Bank of Australia publishes reams of data monthly. Processing all of it, finding the meaningful signals, and filtering out the noise? That’s where algorithms excel. They don’t get bored. They don’t overlook the obscure economic indicator that turns out to matter.

Sentiment Analysis Has Gotten Genuinely Useful

This surprised me. Early attempts at using AI to scan news and social media for market sentiment were garbage. The models couldn’t handle sarcasm, context, or Australian idioms. They’d read “RBA ruins everyone’s day” and classify it as negative without understanding that’s just standard financial media tone.

But the current generation of natural language processing models? They’re actually pretty good. They can distinguish between “AUD weakness” as a temporary technical move versus structural concern. They track central bank communication tone shifts before the market fully prices them in.

Several fintech firms and AI consultants in Sydney are building proprietary models that combine traditional fundamental analysis with machine-processed sentiment data. The results aren’t perfect, but they’re adding edge. When AUD positioning gets extreme and sentiment data shows capitulation, that’s actionable information.

Where the Models Fail

Here’s what AI can’t do: understand genuine paradigm shifts. Machine learning models are backwards-looking by nature. They’re trained on historical data, which makes them brilliant at interpolation and terrible at extrapolation.

The pandemic? AI models trained on pre-2020 data were useless. The RBA’s shift to unconventional monetary policy? Same problem. Any time the underlying regime changes—regulatory shifts, new central bank frameworks, unprecedented geopolitical events—the models struggle because they’ve never seen anything like it in their training data.

This is why the “AI will replace currency analysts” narrative is bunk. You need human judgement to recognise when the rules have changed. The algorithm will keep trading the old playbook until it blows up.

I also remain sceptical of high-frequency AI trading for retail participants. The big institutional players have microsecond advantages, co-located servers, and genuinely sophisticated quantitative teams. Running some off-the-shelf ML model from your laptop isn’t going to compete. You’re bringing a knife to a gunfight, and the knife is blunt.

The Practical Middle Ground

What’s actually working is hybrid approaches. Use AI for what it’s good at—processing vast amounts of data, spotting non-obvious correlations, tracking sentiment shifts—but keep human expertise for context, regime recognition, and risk management.

I run a handful of models now that complement traditional analysis. One tracks commodity currency relationships and flags divergences. Another monitors central bank speech patterns for subtle tone changes. A third analyses options market positioning for asymmetric risk scenarios.

None of them make trading decisions. They’re inputs, not oracles. But they’re valuable inputs. When all your models and your fundamental view align, you’ve got higher conviction. When they diverge, that’s worth investigating.

What’s Coming

The next phase will be interesting. We’re starting to see models that don’t just analyse existing data but generate synthetic scenarios. “What happens to AUD if iron ore falls 20% while US yields spike but Chinese stimulus kicks in?” The model can simulate thousands of variations and probability weight the outcomes.

That’s genuinely useful for risk management. Understanding not just your base case but the full distribution of possible outcomes changes how you think about position sizing and hedging.

But—and this is crucial—garbage in, garbage out still applies. If your model’s trained on data that doesn’t reflect current market structure, or if you’ve overfit to historical patterns that don’t persist, you’re just automating bad analysis. I’ve seen plenty of that too.

The Australian dollar’s always been fun to analyse because it sits at the intersection of so many forces: commodities, China, US dollar trends, domestic monetary policy, global risk sentiment. Adding AI tools to the analytical toolkit makes that complexity more manageable. It doesn’t make it simple. Anyone promising simple answers in forex is lying or deluded.

But if you’re willing to do the work—testing models, understanding their limitations, combining machine analysis with market experience—the tools available now are genuinely better than what we had five years ago. That’s progress, even if it’s not the revolution the hype merchants promised.