Books in Bytes
All 16 learnings · one per slide
The Signal
and the
Noise
Nate Silver, in bytes
16 key learnings on why so many predictions fail, and some don't.
16 learnings · ~40 sec eachSwipe →
Library / Big Ideas & History / The Signal and the Noise
The Signal and the Noise, in bytes
16 key learnings from Nate Silver's book on forecasting, from weather to elections to poker. Each takes about 40 seconds to read. Swipe, use the arrows, or your keyboard.
The Signal
Books in Bytes
Learning 1 · Introduction
Why Predictions Fail
The idea
We have more data than ever, but more information isn't more knowledge; most of it is noise.
Example
Many confident forecasts before 2008 missed the crisis.
Takeaway
Separating signal from noise is the hard part.
The Signal and the Noise · Nate Silverbooksinbytes.in
Books in Bytes
Learning 2
The Ratings Failure
The story
Ratings agencies rated mortgage-backed securities AAA, assuming house prices wouldn't fall everywhere at once.
The lesson
Models built on bad assumptions failed catastrophically.
Takeaway
Question your model's assumptions.
The Signal and the Noise · Nate Silverbooksinbytes.in
Books in Bytes
Learning 3
Foxes Beat Hedgehogs
The idea
Drawing on Tetlock, Silver finds pundits with one big idea predict worse than open-minded 'foxes'.
Example
TV political pundits performed about as well as coin flips.
Takeaway
Be a fox.
The Signal and the Noise · Nate Silverbooksinbytes.in
Books in Bytes
Learning 4
Baseball Forecasting
The story
Silver built PECOTA, a system for forecasting baseball players' performance.
The lesson
Combining statistics with scouts' knowledge works better than either alone.
Takeaway
Blend data and expertise.
The Signal and the Noise · Nate Silverbooksinbytes.in
Books in Bytes
Learning 5
Weather Forecasts Improved
The idea
Weather forecasting has become much more accurate through better models and data, plus human judgement.
Example
Hurricane landfall predictions have improved dramatically.
Takeaway
Forecasting can improve with feedback.
The Signal and the Noise · Nate Silverbooksinbytes.in
Books in Bytes
Learning 6
The Wet Bias
The idea
Commercial weather forecasters sometimes exaggerate the chance of rain to avoid blame.
Example
Forecasting 20% rain when the true chance is 5%.
Takeaway
Incentives distort forecasts.
The Signal and the Noise · Nate Silverbooksinbytes.in
Books in Bytes
Learning 7
Earthquakes
The idea
Earthquakes remain essentially unpredictable in timing, though their frequency follows patterns.
Example
Claims of precise earthquake prediction have failed.
Takeaway
Some things can't be predicted precisely.
The Signal and the Noise · Nate Silverbooksinbytes.in
Books in Bytes
Learning 8
Economic Forecasting
The idea
Economists' forecasts are often overconfident and poor.
Example
Few predicted recessions in advance.
Takeaway
Treat point forecasts with caution.
The Signal and the Noise · Nate Silverbooksinbytes.in
Books in Bytes
Learning 9
Self-Fulfilling Forecasts
The idea
Predictions about epidemics can change behaviour and affect outcomes.
Example
Forecasts of the 2009 swine flu affected how people responded.
Takeaway
Forecasts can change the future.
The Signal and the Noise · Nate Silverbooksinbytes.in
Books in Bytes
Learning 10
Bayes' Theorem
The idea
Start with a prior belief, then update it as new evidence arrives.
Example
Updating the odds of something being true after a test result.
Try this
State your prior before looking at new evidence.
The Signal and the Noise · Nate Silverbooksinbytes.in
Books in Bytes
Learning 11
Chess and Computers
The story
When Deep Blue beat Garry Kasparov in 1997, a move caused by a bug unsettled Kasparov.
The lesson
Humans read meaning into randomness.
Takeaway
Don't overinterpret odd results.
The Signal and the Noise · Nate Silverbooksinbytes.in
Books in Bytes
Learning 12
Poker
The idea
Poker rewards probabilistic thinking, but luck dominates the short term.
Example
Silver played professional poker and saw how skill only shows over many hands.
Takeaway
Judge skill over many trials.
The Signal and the Noise · Nate Silverbooksinbytes.in
Books in Bytes
Learning 13
Markets
The idea
Markets are hard to beat consistently, though bubbles happen.
Example
Stock prices following a random walk most of the time.
Takeaway
Be humble about beating the market.
The Signal and the Noise · Nate Silverbooksinbytes.in
Books in Bytes
Learning 14
Climate Signals
The idea
Climate forecasts involve signal (long-term warming) amid noise (year-to-year variation).
Example
A cold year doesn't disprove a warming trend.
Takeaway
Look at trends, not single points.
The Signal and the Noise · Nate Silverbooksinbytes.in
Books in Bytes
Learning 15
Unknown Unknowns
The idea
Pearl Harbor and 9/11 showed failures to see unfamiliar threats.
Example
Terrorist attacks following a power-law distribution.
Takeaway
Imagine the unfamiliar.
The Signal and the Noise · Nate Silverbooksinbytes.in
Books in Bytes
Learning 16
Beware Overfitting
The idea
Fitting a model too closely to past data makes it worse at predicting the future.
Example
A model that explains every past wiggle but fails tomorrow.
Reflect
Where might you be overfitting your past experience?
The Signal and the Noise · Nate Silverbooksinbytes.in
Books in Bytes
All 16 learnings complete
That’s all 16 learnings.
Good forecasts are probabilistic and updated. What belief should you express as a probability?
Abhay Kumar JaiswalFCA · PMP® · Workday Financeabhayjaiswal.com
Summaries in my own words. Read the book for the full stories.
1 / 18
Summaries are written in my own words for learning. Read The Signal and the Noise by Nate Silver (2012) for the full stories. More Books in Bytes