It is shown by M. Moeller (M. Moeller, Shimura and Teichmüller curves) that there are only 2 Shimura and Teichmüller curves in the moduli space of curves $M_g$, namely, the ones given by $y^4=x(x1)(xt)$ and $y^6=x(x1)(xt)$ in $M_3$ and $M_4$ respectively. I have difficulties in understanding this result, because we know that it is true that families of cyclic covers of $\mathbb{P}^{1}$ branched over $1$ point in $\mathbb{P}^{1}\setminus \{0,1,\infty\}$ (squaretiled surfaces) generate Teichmüller curves in $M_g$ (for example by A. Wright, SCHWARZ TRIANGLE MAPPINGS AND TEICHMüULLER CURVES:ABELIAN SQUARETILED SURFACES). On the other hand, the Shimura curves arising from cyclic covers of the projective line are already classified (By Moonen and other authors) and we know that beside the curves given above there are other examples of Teichmüller curves which, when considered in $A_g$ generate Shimura curves. I must have made a mistake somewhere in understanding this result, but can't figure out where! I would appreciate any help.

1$\begingroup$ Perhaps the point is that $M_g$ is not closed in $A_g$. $\endgroup$ – naf Dec 20 '15 at 7:11

$\begingroup$ I don't understand what you mean. If we have $C\to M_g\xrightarrow{i} A_g$, then since the Teichmüller curves are not compact and $i$ is an immersion, we get an open curve in both $M_g$ and $A_g$. The above Shimura curves are also noncompact, i.e., they have cusps. So I don't see the contradiction! $\endgroup$ – Cyrus Dec 20 '15 at 9:11

$\begingroup$ There could still be points on the (noncompact) Shimura curve that are not in $M_g$. (This is just a suggestion, I am not claiming that this is indeed the explanation.) $\endgroup$ – naf Dec 20 '15 at 11:29
I think the point is that, when you have a family $f:X\to C$, then the induced map $C\to M_g$ is not necessarily one to one. It is not even generically one to one but in reasonably good cases it is generically finite. Now by a STcurve completely inside $M_g$ we mean that this latter map is injective and the image $C\to A_g$ is a Shimura curve. These two conditions together are very restrictive. The families you mentioned are Teichmüller in the sense that the image of $C\to M_g$ is Teichmüller, not that the family completely sits in a Teichmüller curve.