The U.S. Census Bureau publishes data on new home sales on roughly the 17th day of every month (www. census.gov/const/newressales). Data are presented both as seasonally adjusted and in raw form. (Periodically, the Department of Housing changes the seasonal adjustment, which has an effect on the reported number and may introduce a degree of error in this analysis.) The housing market, as represented by this data, is a major source of demand for the lumber market. It goes to reason, then, that the price of lumber is linked to new home sales.
Indeed, in looking at “New home sales & inventory”, which covers 2002 through the present, we’re vaguely reminded of the futures market represented in “Lumber prices”. Whether that correlation is strong enough for one to predict the other, however, will require more extensive analysis.
Today, both predictive directions are relevant — home sales as predictive of lumber prices and vice versa. It is, after all, possible to trade lumber prices or exchange-traded funds (ETFs) based upon the real estate index, home builders, real estate prices, etc. However, as traders, we’re only interested if that relationship is measurable and tradable. To begin such analysis, consider the following questions:
• Is there a correlation between sales of new homes and lumber prices?
• Is there a linear regression equation that could estimate lumber prices based upon new home sales with any confidence?
• Is there a time series adjustment that could project home sales — that is, can this month’s lumber prices predict next month’s home sales?
• Conversely, is there a time series adjustment that could project lumber prices — that is, can this month’s home sales project next month’s lumber price?
Before we embark on such analyses, we must consider whether there is a fundamental reason why this should work. Statisticians are aware correlations are nothing other than a comparison of two sets of data. A positive correlation indicates that as one set of data increases, the other set is correspondingly increasing, all else being equal. It does not mean the first set causes the second set to increase; thus the mathematical axiom “correlation does not imply causation.” All too often, traders make the mistake of confusing correlation with causation. If we have reason to conclude there may be a causal relationship between our data sets, it will give us more confidence in any predictive ability of our results.
In this case, the causality is obvious. Lumber’s primary use is in new home construction. If sales of new homes decrease, builders are less likely to begin new homes and the demand for lumber will decrease. Because it is assumed the supply of lumber is relatively constant, absent some global catastrophe, it logically follows that lumber prices are more dependent upon demand than supply factors.