Short on systematic trading-

Lot of time gets spent on optimisation, indicators and multi factor models, not enough on selection of the right asset class or allocation between them. Any strat can make money only if the market regime is suited for it.

1

In a trending market, the entire set of trend following strategies will make money. Some will make more, some less. In a flat trendless market, none of them will make money. optimisation wont matter. Only allocation matters.

2

As markets mature, the sources of alpha change. In a new market, info asymmetry is the prime driver- mostly pre Internet era. As info diffusion speeds up, everything is priced fairly enough to make any 'arb' no longer worth the risk

3

as markets get more efficient, returns drop, as prices are 'fair'. Bid offer spreads were 1-2%+ even for the most liquid stocks in 1990. Its 0.01% now. If you want the alpha from better research, you can only do so in smallcaps or frontier markets.

4

there are only a certain number of "points" you can extract from a given time series in a specific duration through whatever means- theta/directional trading/ etc. The set of strats that approach the global maxima is a smooth curve. overoptimization to get max makes it fragile

5

if you want higher raw $ returns, increase capital deployed. or for higher % returns, increase leverage- which usually costs more drawdown. the only mitigation is a portfolio of low correlation strategies or same strat on diff assets.

6

only 2 things matter- your leverage and allocation to each strat. scale it up for whats in a suitable regime, scale it down to whats not. the mkt movement is not in your control

how do you decide allocation? monitor the drawdown. not all drawdowns are equal

7

a drawdown where the drawdown is uniform is ideal. it is true noise. a drawdown which lasts for a very long time and no pullbacks to high watermark is a market regime change when your system shouldnt have been trading at all.

8

how do you find these ? backtests. its not a guarantee of future performance, just a record of the past. its not meant to be used to discover the strats' pitfalls- those should be obvious from the design. if you design to catch trends, it'll obviously fail w.o a trend

9

Relying on running dozens of backtests to find one that looks good, is the wrong approach. a backtest should be used to ensure that the strat performance is not too sensitive to parameters. if you change an input a bit and perf changes drastically, its overoptimized.

10

in summary-
-focus on choosing the right timeseries to trade at the right time
-always trade a basket and scale allocations to what is working best
-dont rely on backtests alone
-study the drawdown

/end