Every figure, definition and threshold this video puts on screen, chased to a primary source. The worked examples are illustrative: they demonstrate a mechanism rather than record what any particular market did on any particular day, and they are labelled that way on screen wherever they appear.
RSI was published by J. Welles Wilder in New Concepts in Technical Trading Systems (Trend Research, 1978). Wilder set the default lookback at 14 periods and named 70 as overbought and 30 as oversold. The video uses the 14 period setting and the 30 threshold exactly as Wilder defined them.
Wilder’s smoothing is not a simple moving average: the first average gain and average loss are simple means over the lookback, and every value after that is smoothed, which is why an RSI computed with a plain moving average does not agree with the published one.
The point the video makes about RSI is not about the formula. An oversold reading in an uptrend and an oversold reading in a downtrend are the same number describing two different situations, because the reading measures the size of recent losses relative to recent gains and says nothing about the environment those losses happened in.
VWAP is the cumulative typical price weighted by traded size, divided by cumulative size, and it resets at the start of each session:
typical price = (high + low + close) / 3
VWAP = sum(typical price x volume) / sum(volume)
The weighting by size is the whole difference between VWAP and a moving average of price: a price where a large amount traded pulls the line, and a price where almost nothing traded barely moves it. The typical price rather than the close is what is averaged, because a bar that ran high and closed back at its open did trade all the way up there.
The Nasdaq 100 measures 100 of the largest Nasdaq listed non financial companies. The S and P 500 measures large capitalisation United States companies across sectors, including those same Nasdaq listed names. A large Nasdaq listed company is therefore eligible for both indices at once, which is why the two share a substantial number of their largest constituents and why those shared names carry heavy weight in both.
The exact number of names in both indices at once changes with every index review, so the video shows the overlap without stating a count.
The link between an oil price and an airline share price is real, and this is the size of it. IATA’s 2025 industry outlook put jet fuel at 25.8 per cent of all airline operating costs, on an average price of 86 dollars a barrel and a total industry fuel bill of 236 billion dollars, down from 261 billion dollars in 2024.
“Jet fuel is expected to average $86/barrel in 2025 (well below the $99 average in 2024), translating into a total fuel bill of $236 billion, accounting for 25.8% of all operating costs.”
That is roughly a quarter of the cost base, which is the point the video makes with it: enough for fuel to matter to an airline, nowhere near enough for fuel to be the whole company.
For a Pearson correlation to be distinguishable from zero at the conventional five per cent level, the value it has to reach depends heavily on how many observations produced it. Derived from the t distribution, with r critical = t / sqrt(t squared + degrees of freedom):
| observations | degrees of freedom | critical t | correlation needed |
|---|---|---|---|
| 7 | 5 | 2.571 | 0.754 |
| 200 | 198 | 1.972 | 0.139 |
So a correlation of 0.7 measured across seven observations does not clear the bar, while a correlation of 0.15 measured across two hundred does. A high number from a short window is the expected behaviour of random data, not evidence against it.
The same arithmetic covers the run of seven sessions. A run of seven independent coin flips all landing the same way has probability 1 in 2 to the power 7, which is 1 in 128, or 0.78 per cent. In a long enough stream of sessions, runs of seven appear regularly without anything causing them.
Testing many configurations and keeping the best one inflates the best result even when nothing in the data is real, and the more configurations are tried the worse the problem gets. Bailey, Borwein, Lopez de Prado and Zhu showed that a strong backtest is easy to produce from a relatively small number of alternative strategy configurations, and that the probability a backtest is overfit rises with the number tried. Because the number of configurations tried is almost never reported alongside the result, someone reading only the result cannot tell how much of it is selection.
The general form of the problem is not specific to trading. Ioannidis set out why a research finding is less likely to be true when the study is small, when the effect is small, when many relationships are tested with little preselection, and when there is flexibility in definitions and analysis, all four of which describe a trader scanning windows and settings for a correlation.
Two series can produce a very high correlation coefficient with no connection between them whatsoever. Tyler Vigen’s Spurious Correlations project searched large public datasets for exactly this and published nearly two hundred examples, each with a real Pearson correlation and no mechanism of any kind. It is the standard demonstration that a high correlation, on its own, carries no evidence about causation.
The reef counts, oat milk orders, chair squeaks and pizza sales in this video are invented for the sake of the argument and are shown as such. The behaviour they illustrate, that unrelated series routinely correlate strongly, is the documented one above.
Three separate ideas the video keeps apart, and the reason it keeps them apart:
Correlation measures whether two series moved together over a measured window. It carries no information about which one moved first, and none about whether a trade in either was available at a usable price.
Lead lag asks whether one series reliably moves before the other, which is a different measurement: the correlation is computed at a range of offsets rather than at zero, and what matters is whether a peak away from zero is stable rather than whether the value at zero is large.
Concentration is what correlation between open positions actually tells an account holder. Positions that are highly correlated with each other do not diversify risk, so several positions that all depend on the same market direction carry closer to the risk of one large position in that direction than to the risk of several independent ones. This is the mechanism behind the video’s point about holding several names that all track the same index.