Quantamental FX Trading: Blending Macro and Quantitative Analysis to Trade Currencies

Alpha trading is fundamentally about identifying mispricings and getting on the right side of them.

Source: robotwealth.com

Sounds simple enough. But how does one identify when something is mispriced?

That hinges entirely on where you draw the blue line in the image above. And herein lies the problem… how does a regular person know where something should be trading?

The humble (and nearly always correct) answer is that you can’t realistically form a better view on this than the aggregate market.

But you can identify things that tend to cause pricing distortions – cases where an imbalance in supply and demand causes things to trade away from where they should.

A classic example is ETF rebalancing. An ETF manager trades to keep the fund’s exposures in line with its mandate, not to maximize trading returns. When those trades are big enough, they can move the market in predictable ways.

Nearly all of your alpha trading as an indie trader will be based on effects like these. They don’t rely on you having a better model of fair value than the market; instead, they rely on you identifying predictable flows that create pricing distortions.

But in this article, I want to talk about the other type of alpha trading.

In this approach, you build a model of fair value and trade deviations around it:

• If an asset is trading rich to your model’s fair value, you sell it
• If it’s trading cheaper than your model suggests it should, you buy it

This sort of approach is primarily the domain of professional teams with access to data and the resources to use it effectively.

But we’ll explore the idea of “quantamentals” – combining fundamental data into a pricing model – from the perspective of what you might be able to do as an indie trader.

Don’t be under any illusions – it’s highly unlikely that you can outperform a well-resourced professional team using such an approach. But maybe you can build something that adds value to a broader portfolio. And understanding this approach is useful in its own right.

So let’s dig in.

What’s a Quantamental?

When most people hear “quantamental,” they picture some mystical blend of economic PhD wizardry and machine learning magic.

The reality is much simpler. At least conceptually.

Quantamental trading takes economic fundamentals – things like GDP growth, inflation rates, interest rate expectations – and turns them into systematic, quantifiable signals.

Instead of a human portfolio manager manually digesting dozens of central bank speeches and trying to remember what the Reserve Bank of Australia said three weeks ago, fundamental data is fed into algorithms that can process it consistently.

The output of such an algorithm is typically an estimate of fair value, which is what something should be worth given all the information and assumptions in the model.

It might also be a metric that has some relationship or predictive utility over forward returns.

Here’s an example:

Imagine you want to track whether the European Central Bank is getting more hawkish or dovish over time. A discretionary trader might read every ECB statement and form a subjective view. A quantamental approach would create a scoring system – maybe tracking the frequency of certain words in statements, combined with actual policy changes, plus market-implied rate expectations – and produce a daily “hawkishness score” for the ECB.

One important detail is that this data needs to be point-in-time. You can’t use today’s revised GDP figures when testing the model on past data. You need to use exactly what was known at that moment, without hindsight bias.

This matters because markets react to information as it’s released, not as it’s eventually revised. If you’re testing whether currency moves follow GDP surprises, you need the original GDP release that traders actually saw, not the revised version published six months later with all the revisions.

What you’re really doing is building time-series indicators for macro fundamentals, just like you would for prices. But instead of a moving average of EUR/USD, you might have a moving average of Eurozone economic surprise indexes or German yield curve steepness.

This approach bridges the gap between “I think the dollar should be stronger because the Fed is hawkish” and actually having a systematic way to express that view.

JPM Macrosynergy Quantamentals

J.P. Morgan’s quantamental desk built JPMaQS (J.P. Morgan Macrosynergy Quantamental System) in partnership with a firm called Macrosynergy.

It’s a massive point-in-time database of global macro data.

JPMaQS coverage (number of indicators by region). Source: macrosynergy.com

Picture this: every day, the system generates what’s essentially a “Global FX Scorecard.” Each major currency gets scored on multiple themes – growth momentum, monetary policy stance, external balances, valuation metrics.

The Australian dollar might score in the top quartile for growth momentum (mining boom, strong employment), but poorly on valuation. The British pound might have solid monetary policy credibility, but terrible external balances.

These scores aren’t someone’s opinion. They’re systematically derived from dozens of data series – PMI releases, CPI prints, employment reports, trade balances, and central bank communications.

The edge, if there is one, relies on not everyone processing all available information efficiently. Or at least, your model doing a better job of it than the aggregate market.

Let’s say the model identifies that Canada’s fundamentals are strong (rising commodity prices, hawkish Bank of Canada, strong employment) while the UK’s are weak (stagnant growth, political uncertainty, dovish policy shifts).

That’s not enough to go long CAD/GBP, though.

The key insight is that there’s only a potential trade if CAD/GBP is trading at a dislocation from the model’s estimation of what CAD should be worth in GBP.

Even though Canada’s fundamentals are stronger, if CAD/GBP were trading higher than the model thought it should, you would get short.

Essentially, you’re betting on your model having a better estimate of fair value than the market – a really difficult task.

Where Might the Edge Live?

It’s tempting to think that the edge in such an approach lives in the algorithm – that if you use a really clever machine learning model, you can estimate price better than the broader market.

It makes more sense to focus on the data, though.

You’ll need data for phenomena that genuinely drive returns. But you’ll also need to consider how you use it. Many economic time series, for example, are highly non-stationary, presenting all sorts of challenges.

There’s also the idea of “nowcasting” – using models to estimate the current state of an economic metric in real-time. For example, nowcasting models might predict metrics like GDP growth before official data is released.

These models could ingest everything from retail sales and industrial production to satellite data and news sentiment, learning the relationships that best predict official figures.

While you’re not realistically building such a model as an indie trader, one important takeaway is that the edge isn’t the machine learning algorithm. The edge is understanding which economic relationships actually matter and how their dynamics work.

Tools You Can Actually Use

Now for the practical bit – what can you actually do with this information?

You’re not going to build JPMaQS in your garage, nor is a subscription likely to be cost-effective outside of professional trading.

But you can apply quantamental thinking with publicly available data. A good starting point would be focusing on the same types of information that drive institutional strategies.

Here are some examples.

Implied Interest Rates and Yield Differentials capture a large amount of fundamental information in simple numbers. Currency values are heavily influenced by interest rate expectations, and you can track these in real-time through government bond yields and forward rates.

When UK 2-year yields rise above US 2-year yields, that signals the market expects the Bank of England to be more aggressive than the Federal Reserve.

You can plot these yield spreads on any decent charting platform. The spread between US and Japanese 10-year yields tells you how Japanese growth and inflation expectations compare to American ones.

Monthly US-Japan 10-year yield spread. Source: TradingView

CFTC Positioning Data: Every week, the Commodity Futures Trading Commission publishes detailed breakdowns of futures positions held by different trader categories – commercial hedgers, large speculators, and small speculators.

This is basically a weekly x-ray of how different types of players are positioned in currency futures. When large speculators are heavily positioned, it might signal overcrowding that’s marginally more likely to revert.

The beauty of CFTC data is that it quantifies sentiment. The downside is that because it’s publicly available, everyone else knows about it. It’s not a unique, proprietary source of information (nothing available to you is).

ETF and Fund Flow Data: Given the large amounts of capital managed by ETFs, these can give you another window into where global capital is moving. When investors pile into emerging market bond ETFs, that money has to go somewhere – usually supporting those currencies. When they sell Japanese equity ETFs, that creates selling pressure on the yen.

Many ETF providers publish daily flow data. You can track when money is moving into or out of specific regions or asset classes. Sustained flows will likely impact currencies.

This isn’t perfect. ETF flows are just one piece of the capital flow puzzle.

But they’re a real-time, publicly available proxy for the same “money momentum” that big funds monitor through prime brokerage relationships and institutional flow data.

Practical Implementation Tactics

Handling non-stationary data:

Typically, macroeconomic data needs some special handling to make it useful in our models.

It’s common for macroeconomic time series to be drifty and non-stationary:

US real rates time series. Source: robotwealth.com

The main problem with trying to analyze series like this one is that different historical periods aren’t comparable.

A simple solution is to apply a rolling z-score. Simply take each observation, subtract the N-day mean, and divide by the N-day standard deviation:

Rolling 5-year z-score of real rates. Source: robotwealth.com

This gives a much more stationary time series and makes comparisons between historical periods more meaningful.

Realistic time horizons:

Deviations around a notion of fair value derived from a macroeconomic model aren’t likely to converge in the short term. You’re realistically going to be holding positions for weeks or months.

One important practical implication is that your carry costs can add up. For retail FX accounts, your broker will charge a spread on the swap rates (the interest rate differential) – this has the potential to be a significant cost hurdle.

So do some research on where you can get trading conditions amenable to the sort of hold periods you’ll encounter.

Deciding when a deviation is big enough:

One thing is certain in trading: your costs.

Every time you trade, it costs you money. So you need a big enough deviation from fair value in order to make a particular trade worth your while.

Here’s another certainty: your model of fair value will be wrong.

It’s unlikely that you can perfectly model the price at which something should be trading. A very practical consequence of this is that, in addition to your cost hurdle, you should also include a hurdle to allow for model error.

Essentially, you want to make it worth your while, given both the costs of trading plus some wiggle room for being wrong.

A practical approach to deciding when a deviation is big enough might be:

1. Calculate a z-score of the spread between your model’s fair value and the spot price. This will give you an idea of where the extremes of deviation lie and provide some insight into how price tends to diverge and converge with the model’s fair value. Use this to decide on a threshold for entering trades. Let’s say that’s a z-score of 2.
2. Determine a raw percentage from the spot that makes it worthwhile from both a trading cost and a model error perspective. Let’s say that’s 1%.
3. Enter trades when the z-score of the spread exceeds the threshold and the raw deviation of price from the model fair value exceeds your trading hurdle.
4. Close when the spread converges.

Broader considerations:

Don’t try to trade like a specialist. Your edge as a retail trader is diversification across strategies and markets. You have the flexibility to go wherever the opportunities are strongest.

Instead of building the perfect valuation model, you should be identifying multiple edges across different markets and allocating capital systematically across all of them.

Maybe you’re trading FX carry, basis arbs in crypto, and turn-of-month seasonality in bonds. Each strategy might be noisy on its own, but together they can produce surprisingly stable returns. If you can fit a quantamental model in there, great. But don’t think that it needs to do all the heavy lifting.

Focus on edges that make sense, not techniques that sound impressive. Machine learning and artificial intelligence are tools, not edges. The edge comes from identifying when something is trading cheap or rich relative to where it should be trading.

It’s certainly easier to find situations where price-insensitive trading is likely to create price distortions than it is to build a better model of fair value than the market.

If you choose to do the latter, you should be under no illusions as to where your edge is coming from and how difficult it is.

The institutional players have better data, faster execution, and more sophisticated models. But they also have constraints you don’t have – regulatory requirements, client mandates, career risk considerations that prevent them from taking certain positions or changing strategies quickly.

Your advantage isn’t trying to be a better version of them. Your advantage is being systematically opportunistic across multiple edges, asset classes and time horizons in ways that institutional structures don’t permit.

Focus on understanding why edges exist rather than just identifying that they exist. Build frameworks you can apply consistently rather than getting caught up in the excitement of individual trades.

Most importantly, remember that uncertainty is everywhere in this game. You can’t eliminate it, but you can stack the odds in your favor through good analysis combined with solid reasoning about what you’re trading.