Quant book

Systematic, cross-asset, multi-strategy · Simulated, daily

Loading.


Book today

Each strategy is shown at its own 10% volatility scale. The book holds one ninth of each, times a diversification multiplier. Alpha is what is left of the return after removing the part explained by the S&P 500, per unit of its own risk.


Strategies


Performance

Window
Exhibit 1 Cumulative excess return, rebased to 100, log scale. A dashed vertical line marks the start of the out-of-sample record once it is inside the window.
Exhibit 2 Drawdown from the running peak
Exhibit 3 Each strategy at its own 10% volatility scale, cumulative, rebased to 100. The grey line is 100.
Table Book, monthly returns in percent

Positions

Exhibit 4 Net exposure by asset class, share of capital, positions held through the session

Positions going into the next session. Weights are shares of capital: 0.25 is a long position worth 25% of the book, a negative number is a short. "Overnight" is held from the close to the next open only.


Trades

Changes in position of at least 0.2% of capital over the last ten sessions and into the next one. A position takes effect in the session shown and is traded at the close before it. Reactive positions are left alone until the target leaves a band of 10% of the average size.


Signals

Table Twelve-month momentum is the excess return over its volatility, in standard deviations. Momentum and value ranks are inside the asset class, 0 to 100. Carry is the annual return if prices do not move.
Monitors Read daily, not traded

Risk

Table What the book did in past episodes, against passive portfolios

Volatility, VaR and expected shortfall come from the book's own daily returns over the last two years; VaR and shortfall are one day at 95%. The book is short volatility in more than one sleeve, so its bad days cluster when volatility jumps: the two volatility shocks in the table are its worst episodes.

Table Correlation between strategies, last five years

What the evidence supports

Table Five criteria, fixed before any result was seen
Table Against passive portfolios, full simulated period
Table Sharpe ratio by starting year: does the result depend on where the sample begins?

The chance that the gain is real comes from a block bootstrap of the difference in Sharpe ratios. The deflated Sharpe ratio (Bailey and Lopez de Prado) asks how likely it is that the true Sharpe ratio is above zero once you allow for the number of ideas tried. Trying many ideas and keeping some produces a good-looking backtest by luck alone, and this is the correction for it.

Table Every idea tried, in the book or not

Data


Limits

A backtest is not a track record. The sample starts in 2008. Several strategies are long equities part of the time, so the book has a correlation of about one half with the S&P 500 and part of its return is equity risk taken at chosen moments. It loses sharply when volatility jumps. Effects of this kind can fade once they are widely known, and some depend on trading at very low cost. ETF closing prices are not always tradable in size.

Educational research. Simulated results have inherent limits and do not represent actual trading. Not investment advice.