Presenting a quantitative analysis
Seven slides, one sentence and one number or chart each, in an order that mirrors the work. Plus the questions a careful reader will ask and how to have the answers ready.
A short analysis deserves a short presentation. The structure below mirrors the workflow, so the slides can be started as soon as the baseline number exists and filled in as each stage completes. Stop touching the analysis well before the deadline; the last stretch is slides only.
The rule for every slide
Title, one sentence, one number or one chart. Nothing else. If a slide needs a second number, it is two slides or the second number goes.
Seven slides
| # | Title | The sentence | What is on it |
|---|---|---|---|
| 1 | The question as I read it | Cross-sectional or single-series, horizon, score, baseline. | the four choices |
| 2 | The data, and three things wrong with it | Size and span; the three findings, each with its decision. | three bullets |
| 3 | How I decided what counts | The protocol: split dates, embargo, score, fixed before any result. | the five lines |
| 4 | Baseline result | The simplest signal, scored on the hold-out, read out in words. | the report line(s) |
| 5 | What else I tried | The hypothesis round, on both periods, and what happened. | the report lines side by side |
| 6 | Does it hold, and could you use it | Stability inside the hold-out; turnover and break-even cost. | cumulative-IC chart, cost sweep |
| 7 | What I would do with another day | Three follow-ups, ranked. | three bullets |
Filled in for the reversal study
| # | The sentence |
|---|---|
| 1 | Cross-sectional, one day ahead, mean daily rank IC against a zero baseline. |
| 2 | 483 tickers over five years; survivorship-biased by construction; 22 short histories dropped; extreme moves kept but ranked. |
| 3 | Time split at mid-2016, one-week embargo, hold-out reported once. Fixed before any result. |
| 4 | Out of sample: IC +0.016, t just under 2, positive on 55% of days. Small and real-looking. |
| 5 | Five-day reversal beat it in-sample and lost out of sample. I tried two things; the simpler one held. |
| 6 | The effect sits in the first half of the hold-out; 85% daily turnover; dies at about 2.4 bps round-trip. |
| 7 | Test decay properly with a rolling IC; skip a day to rule out bid-ask bounce; condition on volume. |
The full version, with the numbers behind each sentence, is in the case study.
Charts
Three charts do most of the work and all of them are a few lines of matplotlib saved as PNG: cumulative IC over the hold-out, net return against transaction cost, and the mean outcome by signal decile. IC by year and a histogram of the target with its tails are the next two. See charts as PNGs for slides.
The questions to expect
Have one honest sentence ready for each. These are the ones that come up for any signal-scoring analysis; the answers below are the ones the reversal study supports.
- Why rank correlation? Forty-six rows have moves over 25%; a raw correlation would let one of them decide a day. Ranks make every stock count once. On the hold-out the Pearson IC was +0.006 (t 0.8) against Spearman +0.016 (t 2.0), which is the fat tails doing exactly that. See Pearson vs Spearman.
- Is that significant? t of about 2 over 399 days, having tried two things rather than two hundred. Borderline yes; I would want more data before believing the size of it. See t-statistics from repeated looks.
- Isn’t this just X? (bid-ask bounce, volatility clustering, the market factor.) Name the boring explanation and say how you tested or would test it. For reversal: skip a day between signal and target; done, and the effect vanishes, so bid-ask bounce is at least part of it.
- Why did the second idea fail? It didn’t fail; it looked better in one period and worse in another, which is what noise does with two candidates. That is why the hold-out was fixed first.
- Would you use it? Not at that turnover. Say the break-even cost and who could trade under it.
- What about survivorship? The file has only the survivors, so anything that buys losers is flattered. It cannot be fixed with this file; it would need the point-in-time constituent list.
The honest sentence
Every presentation of an exploratory result should contain a version of this: I tried N things. This one held; those did not. A reader who hears it knows that the number they are looking at was not selected from a hidden pile of worse ones.