Data, Analytics and Coding

A Predictive Model of Antiacid-Driven Wealth Accumulation on the Aviation Trading Floor

This paper presents a rigorous empirical investigation into a business opportunity of considerable magnitude. Specifically, we examine the linear relationship between the aviation trading floor revenues (USD millions) and the projected annual income of Martolda in her capacity as a Gaviscon supplier embedded within said trading floor.

Revenue Up => Stress Up => Acid Reflux Up => Gaviscon Demand Up => Supplier Income Up

This mechanism has been validated through years of direct field observation by the author, who has witnessed colleagues reaching for antiacids the more the team revenues increased.

Martolda modelled the relationship using Ordinary Least Squares (OLS) Linear Regression:
y = B0 + B1*x + epsilon

Where:
- y = projected Gaviscon supplier annual income (USD)
- x = trading floor revenues (USD millions)
- B0 = baseline income (even on quiet days, someone has heartburn)
- B1 = marginal increase in Gaviscon income per additional million USD of revenue
- epsilon = error (days when colleagues switched to Rennies, the traitors)

The dataset was generated from 36 months of synthetic observations. Each data point represents one month of trading floor activity. Gaviscon income is estimated based on unit consumption rates, wholesale pricing, and a 40% markup that the author considers entirely reasonable given the captive nature of the market. LOL :D 

The regression coefficient of approximately $180 per $1M of revenue implies that the trading floor, viewed not as a financial institution but as a gastrointestinal ecosystem, represents a substantially undermonetised commercial opportunity. The author, having spent considerable professional tenure in close proximity to this ecosystem, was uniquely positioned to exploit it. She did not!! :(

A predictive model of Occupational Self-Destruction via Corporate Email Misuse in the Aviation Insurance Sector  

This paper presents a landmark empirical investigation into one of the most under researched threats facing the modern finance professional: the unsolicited “Reply All”. While the industry dedicates considerable resource to modelling catastrophic loss events — hull damage, third-party liability, acts of war — it has conspicuously ignored the single most devastating career event observable on any trading floor. This paper corrects that oversight.

Email Sent => Reply All Clicked => 300 Recipients Notified => HR Involved => Legend Status Achieved  

This mechanism has been validated through years of direct field observation by the author, who has witnessed colleagues achieve immortality on the trading floor through a single misplaced click. The author has also personally contributed to this body of evidence on at least three occasions. She will neither confirm nor deny which three. 

Martolda modelled the probability of a “Reply All“ Catastrophe (RAC) using Logistic Regression:

P(RAC) = 1 / (1 + e^−(β0 + β1X1 + β2X2 + β3X3 + β4X4 + β5X5))

Where:

∙ P(RAC) = probability that the email ends someone’s career

∙ X1 = Thread Length (every reply adds fuel to the fire)

∙ X2 = Confidential Flag (labelling an email “Confidential” is practically an invitation)

∙ X3 = Hour of Day (afternoons are perilous; mornings marginally safer)

∙ X4 = Sender Seniority (MD sends email => junior panics => Reply All)

∙ X5 = Recipient Count (more witnesses = more opportunities for catastrophe)

∙ β0 = baseline log-odds (most people are fine. Most.)

The dataset was generated from 200 synthetic email incidents. Each data point represents one observed corporate email event. Outcome probabilities were calibrated from Martolda's own inbox over the period 2019–2024, which she considers an entirely legitimate and peer-reviewed primary source, heh.

The model achieves a ROC-AUC of 0.896, confirming that "Reply All" Catastrophes are not random acts of fate but entirely predictable consequences of poor judgment, seniority-induced panic, and the word “Confidential”. The Confidential Flag carries the largest positive coefficient, a finding the author considers the single most important contribution to corporate risk management since the invention of the Bcc field. The pseudo R-squared of 0.40 implies that 40% of human email behaviour can be explained by five variables. The remaining 60% remains, for now, unmodelled. The author suspects it is mostly lack of sleep.

An Unsupervised Machine Learning Approach to Martolda's Salary Negotiation Timing on the Aviation Trading Floor

This paper addresses probably the most mysterious subject in the history of humanity after the existence of the Sacred Graal: what is the best moment to ask the CEO (Mr. Nick) for a payrise?

The answer, as with most things in life, depends on several variables — none of which are taught in any business school.

Nick's Mood is High => Mrs. Nick's Mood is High => Nick Had Only 2 Espressos => Bonus Was Recent Enough for Guilt But Not Recent Enough for Complacency => Martolda Gets a Pay Rise

This chain of inference was learned empirically, through a combination of rejected meeting requests, badly-timed corridor convos, and one deeply regrettable attempt of chatting on a rainy Monday morning when Nick didn't have the umbrella.

The dataset comprises 36 months of synthetic observations, each representing one month of trading floor activity. Four variables were selected based on a combination of theoretical relevance, practical observability, and the fact that Martolda's field agent (Olivia the intern) could plausibly monitor them without arousing suspicion:

VARIABLE, DESCRIPTION, SOURCE   

Variable: Nick's mood - Description: on a scale from 1 to 10  - Source: Martolda's direct observation

Variable: Mrs. Nick's mood - Description: from 1 to 10 - Source: classified

Variable: Number of coffees - Description: >4 is biological red flag - Source: Martolda's field agent

Variable: month of Last bonus - Description: month - Source: Payroll intelligence

K-Means was fitted with k=3, Nick's most frequent three states: approachable, negotiable, and do not under any circumstances knock on that door.

Prime moment | Nick mood elevated, espressos at or below 3, Mrs. Nick in positive territory, bonus guilt window active. No half term - kids are at school. Proceed immediately. Bring a one-page summary of your contributions. If rainy day, ascertain the CEO has the umbrella.

Acceptable |  Conditions are tolerable but not ideal. At least one variable is suboptimal. May proceed, but manage expectations. Avoid Mondays, embarrassing convos in the corridors, Martolda's questionable jokes and any day following a compliance meeting. 

Abort mission |  Multiple adverse indicators present. Nick is either over-caffeinated, under-rested, Mrs. Nick is in bad mood. Retreat. Do not make eye contact. Wear non-flamboyant colours. Try again next month.

The Prime moment cluster implies Nick's good mood, Mrs. Nick good mood, no half term,  coffees <=3, ok bonus timing, the umbrella is present. This is more likely to happen in September and October. Please note the author has used this model for the first time in October 2025 to get a payrise in March 2026. She will update you accordingly.

Also, please note that HR department has been attempting to predict Nick's mood for eleven years with no measurable success whatsoever, and they have access to his calendar.

A predictive model of Friday 4.55pm emergencies landing on Martolda's desk, while the whole trading floor suddenly disappears with no trace.

This paper is based on 36-month data and predicts the following 12 (which, in this case, is the year 2025), answering a question the author considers both scientifically important and personally distressing: is it going to get worse? 

Friday arrives → clock approaches 5pm → universe detects imminent broker escape → emergency materialises → colleagues magically disappear → Martolda’s coat goes back on the hook 

The forecast extends the historical trend linearly and overlays the established seasonal pattern, with uncertainty widening over the forecast horizon:

Ŷ(t+h) = T(t+h) + S(t+h) ± 1.96 · σR · √h

Where:

∙ Ŷ(t+h) = forecast number of 4:55pm emergencies at horizon h

∙ T(t+h) = projected trend component (upward; the author already knew!!!!)

∙ S(t+h) = seasonal component (the known monthly rhythm of suffering, repeating faithfully)

∙ σR = standard deviation of historical residuals (the universe’s baseline cruelty)

∙ h = forecast horizon in months (the further ahead, the wider the band)

The lower bound is floored at zero. Negative emergencies have not been observed in the field. The author does not expect this to change.

Colleagues disappearing like smoke as soon as the emergency materialises could be the subject of Martolda's next model. 2 of them said they have recently started suffering from back pain. It must be due to Martolda's voodoo and casting curses.

The point forecast confirms that 2025 will be worse than 2024, consistent with the upward trend identified in the decomposition model. December 2025 carries the highest forecast of any month, with an upper confidence bound of 14.7 emergencies — a figure implying approximately 3.7 emergencies per Friday, meaning the author receives an emergency, resolves it, returns her coat to the hook, and receives another before 5pm. 

The model also highlights a peak of emergencies on 24-Dec and the day right before Bank Holidays.

Martolda refuses to update the model for the year 2026, better not to know.