Saeed

So yes, you can have some blanket rule, for attempting to make time series stationary, but it can often be more involved and techniques like fractional differencing can help. 9/n

The idea is that we can difference time series just enough to make them stationary, but not too much to lose the signal associated with it. 8/n

In @lopezdeprado book (Advances in Machine Learning) he describes the use of fractional differencing. 7/n

If you difference stuff 100 times, the output is going to be stationary.. but it's also going to be noise, any signal is lost. So we need to ask what amount of differencing is appropriate? 6/n

Plus, you have events which produce big outliers so suddenly what was only stationary certainly isn't, in particular looking at COVID. 5/n

There are many tests for stationarity that you can use to understand whether a time series is stationary, but none are perfect, and you can get conflicting answers. 4/n

The simplest approach is for example using first differences (given both quantities are yields) and returns for prices so the dataset "becomes" stationary. 3/n

We use regression models to do inflation forecasting, so it's a super important for us to understand how we preprocess the variables before feeding to the regression. 2/n

On fintwit there's a tweet going round with someone regressing levels of 30Y mortgage rates with 2Y Treasury yields... Typically, when doing such regressions you want to use changes, but the question is what type of changes? 1/n

Tariffs across the supply chain www.ecb.europa.eu/pub/pdf/scpw... #QuantLinkADay

Managing the risks of inflation expectation de-anchoring www.ecb.europa.eu/pub/pdf/scpw... #QuantLinkADay

Thanks for the shoutout! There has been a lot of interest from folks in terms of underlying datasets for CPI (I assume there will be a similar interest for labour data now). Stuff like food and energy inflation, you can capture, if you collect enough data from multiple sources.

So the current approach for BLS labour statistics is to reduce their resources and fire people to “increase accuracy”…. I assume measures like ADP are going to become more important, given they’re independent? Or statistics collected by individual US states?

- This is for direct applications not recruiters etc. 5/n

- Have an understanding of techniques such as cloud based deployment (eg. using ec2, s3 buckets etc.), unit testing etc. 4/n

- Have good knowledge of Python and data science libraries such as Pandas, Scikit-Learn, SciPy etc. and libraries for webapps/visualisation such as Dash and Plotly 3/n

- If you're based in UK/Europe timezone (without short flight away)
- Have at least MSc in a quantiative subject 2/n

We're looking to hire a junior data/software engineer at Turnleaf Analytics to work on our forecasting infrastructure and to do client facing APIs. If you fit the bill in the following tweets and are interested in applying for the role, drop me a DM! 1/n

FX Interventions and Capital-Constrained Banks: Evidence from USD/ILS Spot, Forward, and Option Markets papers.ssrn.com/sol3/papers.... #QuantLinkADay

Your AI, Not Your View: The Bias of LLMs in Investment Analysis arxiv.org/abs/2507.20957 #QuantLinkADay

Details Matter: Loan Pricing and Transmission of Monetary Policy in the Euro Area papers.ssrn.com/sol3/papers.... #QuantLinkADay

Money Talks: How Foreign and Domestic Monetary Policy Communications Move Financial Markets papers.ssrn.com/sol3/papers.... #QuantLinkADay

The International Role of the U.S. Dollar – 2025 Edition www.federalreserve.gov/econres/note... #QuantLinkADay

Re-evaluating Short- and Long-Term Trend Factors in CTA Replication: A Bayesian Graphical Approach arxiv.org/pdf/2507.15876 #QuantLinkADay

Winners!

FinDPO: Financial Sentiment Analysis for Algorithmic Trading through Preference Optimization of LLMs arxiv.org/abs/2507.18417 #QuantLinkADay

Yes it’s a big flow for some countries. I have never tried to model it systematically but there must be some FX impact.

Happy birthday!

Was in Switzerland this week. Amazing nature, but the burgers are expensive (35 CHF) and don’t order all you can eat raclette: you will eat way more cheese than you will have ever eaten in your life.

I reckon I’d have a burger, but a really really good one.