Does JBTD actually work?

Macro Man was pleased to see that others found yesterday's study as interesting as he did, though amusingly not every interpretation was as thorough as his own.  In the interest of completeness, Macro Man thought he would provide a further metric- the daily drawdown from the high to the close, this time using S&P futures to capture price action outside NYSE hours.

If for some reason you were under any illusions about whether the recent uptick in intraday rallies off the lows was an artifact of volatility, the chart above should lay them to rest.   JSTFR has worked just as well as JBTFD over the last several months.   Interestingly, volatility off of the lows seems to lag that off of the highs, as there are a number of cases where the blue line above turns before the orange line.  This can possibly be explained by a tendency to close near daily lows at the beginning of a downtrend and to close near highs at the beginning of an uptrend.   (The fact that the orange line is generally above the blue line is a function of the fact that the SPX tends to close somewhat above the midpoint of the day's range, as discussed yesterday.)

While there are probably more technical insights to be gleaned from a study of intraday price action, Macro Man could not resist the urge to test the proposition, so often expressed in some quarters, that markets are manipulated higher and thus buying dips on a daily basis is "easy money".  Obviously it is, if you look at a chart of yesterday's price action and say "if you bought there (pointing at the lows) and sold there (pointing at the close), you'd have made money."  Ex-ante generally proves to be a little more difficult, as the market has a funny way of not telling you where the low of the day will be ahead of time.

Given that Macro Man is working with a fairly large data set (daily open/high/low/close data for S&P futures since 1982), his approach is somewhat necessarily blunt.  In a fairly interesting coincidence, he observed that the daily average drawdown in Spooz from the previous day's close to the lows was 0.75%, very similar to the average intraday rally from the lows to the close of 0.79%.   (Actually, it's not that interesting at all if you think about it for a second....the average daily close to close change is only a couple of basis points.)

Anyhow, Macro Man looked at two different ways of identifying a "buyable dip".   The first was to buy a dip in line with the long-term average, i.e. a 0.75% decline from the previous session's close.   If the market traded that low, a long position was established and held for the day, getting sold at the close.  If the market never traded that low, there was no position for the day.

The slightly more nuanced approach was to use a vol-based entry point.   Macro Man calculated entry points using a one standard deviation trigger (using a 30 day rolling window)  from the previous day's close.   The more volatile the environment, the bigger a dip there must be to entice the model to buy.  Makes sense, right?  Other than that, the methodology is identical to that described above: once taken, a position is held til the close, then sold.   If there isn't a one SD drawdown on the day, there's no position.

Macro Man ran the model using daily data since 1982, and the results.....weren't good.  The summary stats are set out below.

Here are the equity curves....

...and the annual returns

Where to start?   Clearly 1987 is the elephant in the room thanks to Black Monday- that's the price you pay for not having a risk management filter in your model.  While that early setback clearly impacted the summary statistics, it's not as if the other years represent a strategy that you'd want to put your money behind, either.  FWIW, Macro Man did adjust the parameters in both directions, with little meaningful impact upon the model's risk-adjusted return.

Unsurprisingly, the model does seem to be a bit directional, with notably poor performance during equity bear markets.   However, returns are less uniformly positive during bull markets.   Lest we forget, meanwhile,   Macro Man hasn't included any transactions costs in this analysis; given the frequency of trading, the real returns would be considerably worse than those posted above.

Leaving that issue aside, what would returns look like if we introduced a basic risk management filter to mitigate against the Black Mondays of the world?  Macro Man devised a model with the same entry triggers described above, but with stops at double the magnitude, i.e. a 1.5% and 2 SD drawdown, respectively.  If a stop is triggered, the day's trading is done.   Otherwise, any position is again sold at the close.

The good news is that Black Monday and 1987 are eliminated as outliers.   The bad news is that overall performance, while better, still stinks:

The SD model actually did pretty well during the fantastic bull market of the 1990's.   Since then....not so much.

The annual returns still aren't very good....

So there you have it.    To be sure, some algos no doubt have plenty of mean reversion built into their signals, but that entails selling rallies as much as buying dips, and is likely driven by the sort of order-spamming activity that is anathema to any proper macro punter.  Moreover, seasoned discretionary analysis can almost certainly identify solid entry and exit points for the occasional dip-buying flutter.  But based on a few different approaches, mechanistically buying dips doesn't appear to work now or over longer periods of time.   The best that Macro Man can say for it is that if you're prepared not to use stops, you're probably up a bit this year (per the 4th chart above.)  Then again, the only equity trade that Macro Man can recall referencing this year, buying GDX, is up some 35% since the initial entry point.

In sum, JBTFD JDFW.