To some degree, the PMO (or thermal shock) varies depending on the particular resistor technology and in the case of Vishay parts, the particular model. PMO is done without power generally although some specs do call for it in certain instances. The temperature extremes used also varies depending on the part or technology, it is different for PWW resistors than film/foil. The military thermal shock specifies -55°C to +125°C or +150°C (again depending on the resistor) and sometimes the lower temperature is higher than -55°C. Initially, thermal shock was developed to try and weed out so-called defective resistors, those that may go out of spec at some premature time or may fail due to possible manufacturing flaws. Later, it was applied to resistors for the purpose of 'speeding up' the aging process as such, an attempt to reduce early stage drift (1st year). This 'speeding up' using thermal techniques does not produce substantially predictive results, it does reduce initial aging but not within an accurately predictable range. The number of cycles also can vary, 5 tend to be the usual number but it can be more or less.
Resistor drift as a group is generally predictable based on history of measurements. For example when a data sheet says that a particular resistor type will change by X% because of a certain prescribed condition (such as power, temperature, soldering, ect.), you can be pretty certain that as a group most or nearly all resistors will stay within that percentage. This specification is by actual testing, not mathematical prediction. The problem is that resistor drift is essentially an uncontrollable characteristic, each resistor is going to behave a bit differently than the next when drifting, it is not possible to accurately predict the drift of any one resistor within a group except by chance. Drift as a whole is only specified on the basis of past performance of fairly large quantities of parts. In the case of the so-called 'cheap' resistor, it is not designed to have a low drift characteristic, merely examine the drift specifications in the resistor data sheet and you will see very significant drifting for the same or similar conditions that precision resistors will exhibit much lower drift for. This also means that 'cheap' resistors will generally not develop much lower drift rates with time like the better precisions do.
Trying to use 'cheap' resistors for a precision application is like trying to change the proverbial sow's ear into a silk purse.
One other note, cheap resistors are going to have significantly higher noise levels than the best precisions, PWW resistors do not exhibit 1/f noise for example, virtually all other resistor types do to some degree. PWW and BMF have the lowest overall noise as well and that is another very important characteristic, especially if you're making a high precision Vref.