Fieldnotes by Jacob MackeyLondon, UK · PortfolioGet in touch ↗
Jacob Mackey.
An introduction

Data scientist
in London.

With a background in economics, I build models that predict real-world numbers and test whether they actually work.

You can read my CV, look through my code on GitHub, or email me.

Markets, behaviour
& uncertainty.
Jacob at his desk, working on his computer in a warmly lit room
At the desk.Full-resolution image ↗
01 / In the fieldRight now I’m trying to answer
In progress

Can I predict tomorrow’s electricity prices in Germany?

Every day my model predicts tomorrow’s prices before the market sets them. Each forecast is locked in, then checked against the real prices.

Follow the experiment ↗
Forecast Ledger artwork over wind turbines and electricity transmission lines
Forecast recordPublished before the outcome
Next forecast

Published daily before 12:00 Berlin

Record

Loading the latest public record…

Public record · refreshes on load
Final verdict

31 Oct 2026

Forecast → locked → actual → scorePublic record ↗
The latest scored dayForecast vs actual
ForecastLikely rangeActual pricePrice in € per MWh · Berlin time

Loading the public forecast record…

Source: public forecast ledger ↗ · figures refresh on load

Forecast

The line shows what my model predicted the price would be for each hour, using weather forecasts for wind and sunlight plus recent electricity prices.

Likely range

The shaded area shows the range of prices my model thought was plausible.

Actual price

The dots show the price the market actually set for each hour.

Field study / WindThe forecast, in motion
German wind forecastingInside a 3.4 MW wind turbine
Wind over Germany, live
—m/s, 100 m up

Loading the wind reading…

Averaged over the eight places my forecast reads · Open-Meteo ↗

Inside the machine
the forecast describes.

Scroll to take it apart; hover a part to name it.

—
—

Loading the turbine study…

The rotor follows a capacity-weighted average of 100 m wind readings at the forecast’s eight German locations. The live reading refreshes every 15 minutes; the model shows the corresponding rotor speed.

Wind forecasting study ↗
02 / The work index14 projects · 5 fields
A working principle

Different subjects, one habit: being honest about what I don’t know.

01

Markets

3 projects
01

Can I predict tomorrow’s electricity prices in Germany?

Forecasts recorded before the auction and checked against the outcome. The evaluation is still running.

Live research →02

When do German power prices turn negative?

In 6.2% of the observed hours, concentrated in periods of low demand after wind and solar generation.

Code on GitHub ↗03

Can machine learning improve Germany’s public wind forecast?

Not reliably, so I recommended keeping the official forecast.

Completed study →
02

Finance and trading

4 projects
Portfolio risk study

What happens
in the worst fall?

Worst portfolio fall in the study
Crypto portfolio
69%
With some cash
54%

From the Bitcoin and Ether portfolio study below.

04

Can order-book data predict the next price move?

Slightly, but the moves are smaller than the cost of trading them.

Code on GitHub ↗05

Does rebalancing make a crypto portfolio safer?

Barely, because Bitcoin and Ether crash together. Only holding some cash cut the worst fall, from 69% to 54%.

Code on GitHub ↗06

Is the classic crypto futures trade still worth it?

Barely. It paid 5.1% a year against 4.6% from Treasury bills, and the extra has mostly gone.

Code on GitHub ↗07

Can a company be worth something and still run out of money?

Yes. In this made-up case, a £77m business still can’t repay its 2028 debt.

Code on GitHub ↗
03

Operations and decision-making

4 projects
Retail journeys2.35%

reached a basket

View1,404,179
Basket32,934
Transaction9,952
08

How can AI help a complaints team without making the final decision?

A review queue, proposed routes and evidence-backed summaries, with a person responsible for the decision.

Prototype · case study →09

How should a sports centre review its hire prices?

Completed-year bookings and operating constraints supported holding most prices.

Workplace study →10

How suspicious should a payment be before it’s flagged?

Suspicious enough that checking it costs less than the fraud it stops. 57 of its 69 alerts were real.

Code on GitHub ↗11

Where does the observed shopping journey stop?

Only 2.35% of product-viewing visitors had a subsequent basket event in the dataset.

Completed study →
04

Applied AI

1 project
0102030405060708
Eight synthetic investigations
Retained reasoning · compression
05

Products

2 projects
Field study / TransmissionFrom generation to the grid
Grid transmissionA 380,000-volt power line
−8° elevation

Every forecast ends up
on a line like this.

Supply has to match demand, moment to moment.

Loading the transmission study…

Scroll to follow the camera around the original transmission corridor.

Forecast Ledger ↗
03 / A little backgroundLondon · open to UAE relocation
Portrait of Jacob Mackey
Jacob Mackey
Data scientist · financial economist
About

From economics to data science.

I studied economics and finance. Now I build tools that help with real decisions, check they beat a simple guess, and are honest about their limits.

2024

MSc Financial Economics (Merit)

Cardiff University

2023

BSc (Econ) Economics & Finance

Cardiff University

Tools I reach for

Python · SQL · DuckDB · scikit-learn · XGBoost · Responses API

Outside the work

At the gym.

The gym is a backbone of my daily routine.

I built Overload, the workout tracker I use every day and keep improving. It’s available on the App Store.

Overload on the App Store

Filmmaking is another hobby. This is a short edit I made of a trip to the gym.

A gym session, edited by me · 45 secOpen film ↗