The Federal Reserve came out and said that they should stop stimulus in the form of QE, which lightly shook markets up. It now takes 1 more yen to buy a dollar, and most currencies seemed to have a 0.6-1% swing weaker against the dollar, but little is seen so far in the equity markets. Sure the Nikkei is up a bit and most others down-to-flat from the Fed decision, but nothing world-changing. This too shall be shaken off, and given the rather spectacular rallies this week across most global indices I would have expected a greater pullback and volatility than what amounts to a shrug of the shoulders.
More interesting instead is revisiting Hong Kong non-bank financial institution (NBFI) liquidity that is sponsored by the HKMA. I tested this data for co integration with the HSI last night and got some great results on this chart:
- The series are most likely cointegrated, with a p-value of still having a unit root in regression residuals of 1.584%.
- The relationship between the non-bank OEFBNs and the HSI is negative as the chart yesterday would describe, with a fall of 13.92 HSI points for every HK$ billion of OEFBNs that non-bank financial institutions hold.
- The regression model is rather good, but given the stationarity of the individual factors spurious regression is too much of a worry with an adjusted R^2 of 0.2182.
- The cointegration lag order term is 8, indicating that it takes eight business days for changes in the non-bank OEFBN value to filter through to the HSI.
- Individual differenced series have less unit root probability so day-trading on this might be problematic, but over a longer period of time this should not be a fundamental concern.
This indicates that short-term, consistent changes in NBFI liquidity only goes out to eight days for prediction power over the HSI. If it influences trends in other ways, daily changes are not time-consistent in their impact and thus become very difficult to model. My interpretation would be that NBFIs can thus transact with relatively active traders or investors – of whom a majority follow the trend – for up to eight days before the market reverses the trend.
For those not familiar, below is a quick recap of cointegration methodology. In addition, I will also note a few technical things about the HSI so far today and yesterday.
This sounds horribly mathematical if you’re not mathematically aligned and dreadfully complicated if you have a math background but no economics/econometrics. Relax: it’s basically an autocorrelation test where you want to do this until the autocorrelation isn’t 1 to some variable in the past. Steps for a two-variable model:
- Test the time series you want to use for stationarity, or a fancy word for having a stable expected mean if you use different sampling methodologies. This is most often done with the Augmented Dickey-Fuller test and has been so here as well.
- Dickey-Fuller basically tests the current value against the prior period, augmented Dickey-Fuller assumes that significance falls and generally stops testing time periods after a currently tested period has proven insignificant. If any value has a regression coefficient of 1 to any statistically significant degree, the series is said to have a unit root.
- If the time series are non-stationary (having unit roots), difference the time series by subtracting all data points by the immediately preceding data point. (The original time series is now considered to be integrated of one order higher than the resultant time series.) Repeat this process until you have a stationary series for all data, which we call I(0), or integrated of order zero.
- Take the I(1) series and regress the variables against each other.
- Take the regression errors and test them for unit roots.
- If there is a unit root, the series are not cointegrated.
- If there is not a unit root, there is a likely strong statistical relationship between the variables on the longer term.
- Done! If you have three or more variables and a theory to back their use up, use Johansen co-integration for steps 3 and 4 instead.
Hang Seng Index:
Yesterday was virtually a cavalcade of different positive trade signals which are fading at the moment of writing:
- High volume (HK$79.9 billion) on a massive gap-up day, and virtually all of that volume driving the market higher.
- Crushing the 126 SMA.
- RSI at above 50.
- ADX/DMI positive blowout and the DMI- lower than the ADX trend.
- Turn in the 21 SMA.
Currently, the market is down, testing the 126 SMA at around 23 650, with a retreating RSI attacking the 50 level from above. Today’s trade is also giving high-volume price action on the downside from yesterday’s trading, but it should probably end up a bit lower at ~HK$70-75 billion if this current turnover profile follows the models.
However, in 2-3 more trading days we get a MACD[21, 63, 21] positive histogram divergence, and the 21 SMA will lend support as will the ADX/DMI positive gap. This could be a little bit of a hook pattern where money is redistributed and allows for a northwards Bollinger band surf. Reposition through this week and then go in again next week maybe?