The Research HUB: FX Primer, Pt 5
The primer for understanding and trading FX
Hey everyone,
This is Part 5 of the FX Primer!
5-Part FX Primer Breakdown:
Part 1: FX - Resources, The Big Picture, Variables, Aggregating Knowledge, and Essential Tools.
Part 2: FX - Synthesizing Information from Part 1: Theory, Practice, Causal Mechanics vs. Regression Analysis.
Part 3: Delving into Historical Case Studies: The Importance of Studying History, Continuity vs. Discontinuity, and the Challenges of Backtesting in FX.
Part 5: Integrating Knowledge: Top-Down and Bottom-Up Analysis, Attribution Analysis, the Expectations vs. Actual Matrix, and Quantitative Models.
Intro:
We have finally made it to the last part of the FX primer!
My goal has been to cover each part of the FX analysis process, from identifying the structural regime all the way to understanding how GIP impulses impact returns and risk premiums. The ultimate task is to connect all these dynamics with price action and execution. In my mind, I always aim for a seamless causal chain between all parts of my processes. That is the purpose of this article.
A quick side note on building models: A number of you have reached out asking for examples of FX models or how to build them. Let me provide several principles and thoughts on this:
A model doesn’t necessarily need to be in Excel or code. A model can be a series of functions that you stack, leading you to a specific action. For example, I can create three “if/then” statements:
a: If headline and core CPI are accelerating YoY, then = a bearish bond regime.
b: If the forward curve is pricing an increasing rate of hikes, then = bearish bond regime.
c: If the breakeven curve is upward sloping, then = bearish bond regime.
Then I create an additional function that says, if a, b, and c all = bearish bond regime, then short bonds on a moving average crossover with a 3 stdv stop loss. From here, you can begin to backtest your ideas and see where you can refine them. I basically have spreadsheets filled with very specific and detailed functions like this. I also have many things coded in Python.
Much of this process of model building is about understanding the moving parts, extracting signals from them, and then stacking those signals with functions so that you can align as many things as possible with your execution.
Having all these things in Excel or code is just an added bonus. In my opinion, it's more important to start with the functions and actually know WHAT they mean and WHY they are moving than starting by backtesting a bunch of potential rules.
Alright, let’s delve into some more specifics...
Top-Down and Bottom-Up Analysis:
When I think about top-down and bottom-up analysis, I consider having a process that accounts for all variables.
Top-Down Questions:
How have I quantified the structural regime, and how does this set the probable constraints for the cyclical regime?
How does the current structural regime have continuity and discontinuity with history?
How does the current cyclical regime have continuity and discontinuity with history?
How does the current collocation of structural and cyclical have continuity and discontinuity with history?
How do these regimes create a skew for a specific FX pair?
Bottom-Up Questions:
What are the idiosyncratic variables and drivers that are unlikely to be identified by the top-down regime?
How is the correlation of the FX pair changing against other assets?
How is positioning capitulating or shifting around changes in information?
One thing that is incredibly difficult to backtest is tracking in real-time how information moves across the spectrum from uncertainty to certainty. I essentially try to think about what preconditions exist and how information moves across this spectrum. See the notes in this article for more on this idea:
Attribution Analysis:
Accounting for all the top-down and bottom-up factors allows you to correctly perform attribution analysis on the FX pair.
Here's the thing with attribution analysis: even if you don’t have a clear view of the future, you should always have a reasonable understanding of what is driving FX moves. Regime shifts take place when the underlying attribution analysis shifts, and this is what usually catches positioning offside. As a result, implied vol spikes, and nonlinear moves take place.
Positioning:
This brings us to the whole idea of positioning.
“Positioning” occurs across every timeframe. For example, an entire country can be implicitly long or short inflation due to their underlying balance sheet and income generators.
When we are specifically talking about FX price action, POSITIONING IS ALWAYS IN THE PRICE!
Everyone has all these surveys and COT reports, but in reality, these types of metrics don’t have any predictability for FX. Positioning is reflected when people are using actual dollars to express their views. For example, implied volatility is a great representation of positioning because the premium on vol is priced by market participants in the options market.
These ideas of positioning and attribution analysis are inherently connected because you can begin to see how market participants are acting as the attribution changes. You can also map the degrees to which changes are taking place in GIP and compare them to the size of moves occurring in price.
Another way to identify positioning is simply in correlations. For example, there have been multiple times in past articles and tweets where I have said duration positioning continues to be connected to FX positioning. Here is a chart of bonds and DXY (inverted):
All these things concerning positioning and GIP connect because at inflection points correlations change, and usually, implied vol spikes because participants are caught offsides.
Expectations vs. Actual Matrix:
After we have developed a specific view and identified positioning, then we can look at the calendar for relevant catalysts that are likely to move price action. This is where Substacks by people like
, or going over the week ahead can be incredibly valuable.When I think about an economic data print (or any catalyst providing information), there are several scenarios that can take place.
The data can come out:
Above Expectations
In line With Expectations
Below Expectations
The other scenarios are for the response of price action:
Price moves up
Price doesn’t move
Price moves down
The data vs. expectations can be mapped on varying degrees, and the price action can be mapped on varying degrees.
When I think about data being released and it getting priced into markets, there are two major things I’ll share:
First, it is a test of the underlying thesis. A change in the regime will usually happen incrementally, so all of these short-term catalysts are “tests” to confirm or falsify the view you have.
Second, these are opportunities to see positioning and establish positions. For example, in the most recent macro report (link), I shared that I had a bearish bond view. We then had an NFP print and CPI print that caused bonds to spike marginally. Now, part of the market is interpreting if these economic releases and spikes in price action should be faded or are likely to have some duration.
You can begin to model how economic data gets priced into the market and if you should use the responses to these data prints as entries to trades. This is what brings us to the final portion of quantitative models.
Quantitative Models:
Connecting to the ideas above, ideally, I would like my quantitative models to tell me to short at the same time a data print provides me an opportunity. Things don’t always line up, though, so you have to be careful about over-optimizing.
Big picture, if you have a specific view and a way of seeing how data gets priced into the market, we need some process for quantifying the actual price that provides an additional layer of edge in our favor.
When I think about quantitative models, I categorize them in two major ways:
Momentum: This principle basically says positive returns follow positive returns and negative returns follow negative returns. This is where CTA and trend-type strategies come into play.
Mean Reversion: This principle essentially posits that price reverts to the mean as opposed to continuing in a specific direction.
I am being a bit reductionistic here, but there is a ton of great academic literature on these dynamics in markets. When I think about price, I consider its left tail and right tail exhibiting either momentum or mean reversion characteristics.
Ideally, you want to have some type of combination of the two so you have uncorrelated strategies.
For example, a mean reversion strategy would have a higher degree of success during a period of time like after the initial SVB spike:
Inversely, after this period of time, we have had momentum skewed to the downside:
There are a lot of different ways to map these types of strategies onto your ideas. For example, there are basic moving average crossover strategies in Tradingview that you can use:
There are also basic momentum strategies:
There are Bollinger Bands or ATR Strategies:
The question you need to ask yourself is: what are the specific constraints and goals you are operating under? After you have established your constraints and goals, you can then modify any specific quantitative metrics and risk profile to suit your situation.
I personally run strategies that combine momentum and mean reversion on multiple timeframes and scale position sizing. I don’t specifically use the Tradingview strategies noted above, but stdv and momentum are important inputs. OHLC levels across multiple timeframes (daily, weekly, monthly) are important as well. These have taken me a long time to understand and build, but grasping the principles is incredibly valuable.
After reading more academic papers and books than I care to admit, I will say that it really comes down to how you bring together and weigh the collocation of variables we have discussed.
There is so much opportunity for these types of strategies, especially when you have some type of informational edge or simply a different risk tolerance/timeframe than the market.
Conclusion:
I have provided an overview of the various components of FX. I wrote this FX series with my former self in mind. I wish someone had provided me with a list of all the things I needed to know when approaching FX. I had to go through a lot of trial and error to get to where I am today. That is the process for all of us, though.
I am still improving and refining my knowledge. There are still really significant models I am building that delve even deeper than what we have covered thus far. I started by saying FX is one of the most challenging assets in markets, and now you see why.
I would challenge each of you to quantify and refine the way in which you are executing in your specific domain. Not everyone will be actively building FX models, but the principles and framework in this 5-part series can serve as a mental model for how to think about any domain.
I want to end by saying thank you to each one of you. I truly appreciate the support and interaction. As always, please feel free to reach out anytime via Twitter DM or by email: capitalflowsresearch@gmail.com.
In the information age, you simply need to be at the right place, at the right time, with the right information to succeed
Thanks for reading!